← enghyu한국어 아카이브

AX Trend

AI/AX trends, analyzed from a developer's perspective.

OpenAI Chip Design: AMD GPU RTL Gains Up to 14% PPA
#4055 November 20, 2026 OpenAI Chip Design: AMD GPU RTL Gains Up to 14% PPA OpenAI says its models rewrote AMD flagship GPU RTL for up to 14% better PPA and cut Jalapeno cell area by 13% and 10%, with changes already sent to
Gemini Computer Use on Mac: Release, Controls and Limits
#4065 November 19, 2026 Gemini Computer Use on Mac: Release, Controls and Limits Google brought computer use to the Gemini app for Mac on November 19, 2026. It works in an isolated background window with app permissions and approvals.
Eclipse Enclave 1.0 Release: Sandbox for Coding Agents
#4062 November 19, 2026 Eclipse Enclave 1.0 Release: Sandbox for Coding Agents Eclipse Enclave 1.0 is an MIT-licensed open source sandbox that runs Claude Code, Codex and OpenCode in isolated containers with network allowlists.
Transformers MoE Streaming: 180B on 32GB, 13.9 tok/s
#4057 November 19, 2026 Transformers MoE Streaming: 180B on 32GB, 13.9 tok/s Transformers now streams MoE experts from SSD: a 180B Qwen model runs at 13.9 tok/s on a 32 GB Mac M5, within 6% of Strata. How the slot pool works.
Palo Alto Applied Compute Deal: $4.5B Pricing and Security
#4056 November 19, 2026 Palo Alto Applied Compute Deal: $4.5B Pricing and Security Palo Alto Networks agreed to acquire Applied Compute for $4.5 billion in stock, adding open-weight post-training and inference with security built in.
Claude Code v2.1.328: Auto Mode on Bedrock and Foundry
#4054 November 19, 2026 Claude Code v2.1.328: Auto Mode on Bedrock and Foundry Claude Code v2.1.328 brings auto mode for Opus 4.6 and Sonnet 4.6 to Bedrock, Google Cloud Agent Platform and Foundry, and makes it the default start mode.
OpenAI Jagged Frontier: Coding Agent Success Rates and Costs
#4053 November 19, 2026 OpenAI Jagged Frontier: Coding Agent Success Rates and Costs OpenAI's jagged frontier report: unattended coding agent success on 2 to 4 hour tasks rose from 23% to 46%, with monitoring at 79% and serving code at 25%.
Anthropic IPO Pricing: $83 a Share, Record $81.31B Raise
#4049 November 19, 2026 Anthropic IPO Pricing: $83 a Share, Record $81.31B Raise Anthropic priced its IPO at $83 a share, above the $70 to $78 range, raising $81.31B in the biggest stock market debut on record, per Reuters.
Gemini 4 Neon Release: Pricing, Coding Scores and Limits
#4048 November 19, 2026 Gemini 4 Neon Release: Pricing, Coding Scores and Limits Google Gemini 4 Neon costs $3/$15 per million tokens, tops the Coding Agent Index at 58.6, and cuts cost per task 32% versus Carbon at high.
Lastbook: Offline Survival Library With Local AI and Maps
#4047 November 19, 2026 Lastbook: Offline Survival Library With Local AI and Maps Lastbook 1.1.12 puts manuals, videos, encyclopedias and an offline map on one computer, with optional local AI that answers from the library and cites sources.
Google Stitch Loop Release: 24/7 Product Agent, Waitlist
#4044 November 19, 2026 Google Stitch Loop Release: 24/7 Product Agent, Waitlist Google Labs Stitch Loop is a 24/7 product agent tied to your codebase and live metrics. It opened a global waitlist on Nov 18, with AI Ultra priority.
Suno Visuals Release: 15-Second Music Videos, Pricing
#4042 November 19, 2026 Suno Visuals Release: 15-Second Music Videos, Pricing Suno Visuals, launched Nov 17, 2026, turns library songs into 15-second lyric videos and album art on H3 Move, with Visual credits sold separately.
Apple FastFrames Release: 36ms Video, How It Works
#4040 November 19, 2026 Apple FastFrames Release: 36ms Video, How It Works Apple FastFrames runs 480p video at 36 ms per frame on one B200. The profiling finding, both models, setup steps and limits, explained.
DaisyRose-27B Release: Benchmarks, Training and Limits
#4063 November 18, 2026 DaisyRose-27B Release: Benchmarks, Training and Limits RadixArk DaisyRose-27B doubles Terminal-Bench 4.1 to 12.0% with about 1,200 RL tasks on Qwen3.8-27B. Benchmarks, training cost and limits.
NVIDIA RTX Spark Compact Desktops: Specs, Pricing, 128GB
#4043 November 18, 2026 NVIDIA RTX Spark Compact Desktops: Specs, Pricing, 128GB RTX Spark compact desktops from ASUS, Dell and MSI are on sale now, with 128GB unified memory, a 20-core Grace CPU and up to 1 petaflop claimed.
rust-anything Release: Rewrite Apps in Rust, Piece by Piece
#4041 November 18, 2026 rust-anything Release: Rewrite Apps in Rust, Piece by Piece rust-anything, released Nov 16, 2026, records an app, replays cases against a Rust port and falls back to the original until the port passes.
Proton Lumo Desktop: Connectors, Skills and Permissions
#4038 November 18, 2026 Proton Lumo Desktop: Connectors, Skills and Permissions Proton launched Lumo Desktop on Nov 17, 2026 for Windows, macOS and Linux, adding Connectors, Schedules and Skills with per-action permissions.
Perplexity Orchestrator API: 85+ Models, One Bill, Failover
#4037 November 18, 2026 Perplexity Orchestrator API: 85+ Models, One Bill, Failover Perplexity launched Orchestrator on Nov 17, 2026: one API for 85+ models, unified billing with no platform fee, and same-model failover across providers.
Copilot Autopilot Release: Identity, Memory and How It Works
#4036 November 18, 2026 Copilot Autopilot Release: Identity, Memory and How It Works Microsoft Copilot Autopilot began Frontier rollout on Nov 17, 2026, with its own Entra ID and memory. Home and Office in Copilot are now generally available.
Gemini Tasks Goes Free in Six Countries: Calls and Payments
#4033 November 18, 2026 Gemini Tasks Goes Free in Six Countries: Calls and Payments Google rolled out Gemini Tasks free in the US, Japan, Korea, Canada, Australia and India, with phone booking and AP2 payments starting in the US.
Microsoft MAI Models Beyond Code: Pricing and Performance
#4032 November 18, 2026 Microsoft MAI Models Beyond Code: Pricing and Performance Microsoft expanded MAI beyond code on Nov 17, 2026: Cyber 2 Flash at 92.3%, Image 2.7 ranked No. 1, Realtime 1 from $0.01 per minute.
MAI Code 2 in GitHub Copilot: Pricing and Performance
#4023 November 18, 2026 MAI Code 2 in GitHub Copilot: Pricing and Performance Microsoft launched MAI Code 2 and MAI Code 2 Flash in GitHub Copilot, claiming 98% of GPT-6.1 Sol performance at 40% of the cost on its own benchmarks.
GPT-6.2 Astra and Sol: Release, Benchmarks and API Pricing
#4022 November 18, 2026 GPT-6.2 Astra and Sol: Release, Benchmarks and API Pricing OpenAI released GPT-6.2 Astra and Sol on November 17, 2026: a new Agents Last Exam record, Astra API input and output prices cut 20%, and Sol at unchanged
Olala mini 7A1B Release: OVHcloud First Open LLM, 8GB GPU
#4019 November 17, 2026 Olala mini 7A1B Release: OVHcloud First Open LLM, 8GB GPU OVHcloud announced Olala mini 7A1B, a 7B open model with 1B active parameters that runs on an 8GB GPU and already serves Qwant search answers.
NVIDIA Vera Rubin MLPerf Training: 2x Per GPU Results
#4010 November 17, 2026 NVIDIA Vera Rubin MLPerf Training: 2x Per GPU Results Vera Rubin NVL72 debuted in MLPerf Training v6.1 with the fastest time on six of seven benchmarks and up to 2x GB300 performance per GPU.
Traces Joins Vercel: Coding Agent Sessions and Sharing
#4009 November 17, 2026 Traces Joins Vercel: Coding Agent Sessions and Sharing Traces is joining Vercel with its whole team. The session sharing service supports 16 coding agents and nearly 500,000 shared sessions, with no day one change
Hugging Face OpenRouter Routing: Pricing and Failover
#4008 November 17, 2026 Hugging Face OpenRouter Routing: Pricing and Failover Hugging Face switched language model routing to OpenRouter on November 16, 2026. Requests now pick among 80+ providers live and retry on failure, with no
Claude Code iPhone Duo Simulator: Public Beta Guide
#4007 November 17, 2026 Claude Code iPhone Duo Simulator: Public Beta Guide Anthropic opened a public beta that adds an iPhone Duo simulator to Claude Code desktop, letting Claude fold the device and check layouts.
kosem: Coding Agents Get microVMs, Short-Lived Credentials
#4006 November 17, 2026 kosem: Coding Agents Get microVMs, Short-Lived Credentials kosem gives each coding agent its own microVM, identity and short-lived credentials for GitHub, AWS and Google Cloud. It is in private beta, with pricing
Blender AI MCP Addon: Visual Chat, 21 Tools, Agent Teams
#4003 November 17, 2026 Blender AI MCP Addon: Visual Chat, 21 Tools, Agent Teams mcp-for-blender-pro puts visual chat and a multi-agent team inside Blender, with 21 scene tools and 2,179 stars in four days. Windows x64 limits apply.
OpenAI Healthcare Agents: 99% Safe, 94.3% Bench Score
#4002 November 17, 2026 OpenAI Healthcare Agents: 99% Safe, 94.3% Bench Score OpenAI reports 99% of 325 EHR tasks rated clinically safe, a 94.3% HealthAdminBench score, and Luna at 8.5 cents per task. What the numbers do and do not prove.
Ask AdMob Release: Agent Skills, MCP Server and How It Works
#4001 November 17, 2026 Ask AdMob Release: Agent Skills, MCP Server and How It Works Google Ask AdMob puts a Gemini agent in your account. Agent skills are live today, the MCP server is in early testing, and every change needs approval.
Browser Use 1.0.6 Release: Default Model, 15x Pricing
#4000 November 17, 2026 Browser Use 1.0.6 Release: Default Model, 15x Pricing Browser Use 1.0.6 makes bu-4.0 the default model at $1.80 and $9.00 per 1M tokens, 15x bu-fast. Pin llm='bu-fast' to keep the old price.
Qwen4-Max Release: Pricing, 1M Context and Task Costs
#3998 November 17, 2026 Qwen4-Max Release: Pricing, 1M Context and Task Costs Alibaba Qwen4-Max is out at $1.50 input and $4.50 output per million tokens, with 1M context and a claimed 62% lower average task cost on Work360.
WorldDraft Benchmark: Can Video Models Just Render?
#3973 November 17, 2026 WorldDraft Benchmark: Can Video Models Just Render? WorldDraft tests 12 video models on 432 paired cases across 48 code-built 4D worlds. Conflicting draft cues caused source leakage up to 64 percent.
Claude Code v2.1.326 Release: Code Mode and Agent Teams
#4018 November 16, 2026 Claude Code v2.1.326 Release: Code Mode and Agent Teams Claude Code v2.1.326 lets Code Mode run with agent teams, starts unset sessions in auto mode, and tightens sandbox writes.
Google Ask Maps: Itineraries, Route Tips and Phone Booking
#4015 November 16, 2026 Google Ask Maps: Itineraries, Route Tips and Phone Booking Google added itineraries, route tips, phone booking and saved list search to Ask Maps on November 16, 2026, expanding it to 30 European countries.
Tencent Hy4 flash Release: Pricing, Specs and How It Works
#3999 November 16, 2026 Tencent Hy4 flash Release: Pricing, Specs and How It Works Tencent Hy4 flash is a 224B model with 12B active per token, a 1M context and API pricing far below the flagship Hy4. Specs, pricing and limits inside.
agentes.chat: Open-Source Slack Where AI Agents Are Members
#3997 November 16, 2026 agentes.chat: Open-Source Slack Where AI Agents Are Members Open-source agentes.chat puts AI agents inside team chat as channel members with roles, budgets and approvals. Self-host free under AGPL-3.0 or use the paid
Yogurt Meeting Notes: Bot-Free, Local AI, How It Works
#3996 November 16, 2026 Yogurt Meeting Notes: Bot-Free, Local AI, How It Works Yogurt by FrigadeHQ records meetings without a bot and transcribes and summarizes locally with Whisper and open-weight models. Free, MIT, for Mac, Windows and
Claude Code v2.1.325: Sandbox Permissions and Security
#3995 November 16, 2026 Claude Code v2.1.325: Sandbox Permissions and Security Claude Code v2.1.325 makes sandboxed Bash ask by default, fixes zstd decoding, and closes a compound-command permission gap.
Anthropic Always-On Agents Guide: Five Primitives Explained
#3993 November 16, 2026 Anthropic Always-On Agents Guide: Five Primitives Explained Anthropic's Nov 13 guide breaks always-on agents into five primitives: event inbox, timers, memory, permissions and outbox, plus a 78% cost claim.
Devin Apps Release: Hosted Web Apps on Devin's Machine
#3990 November 16, 2026 Devin Apps Release: Hosted Web Apps on Devin's Machine Cognition's Devin Apps serve the web apps Devin builds from its own machine at a stable URL, with private sharing, 5-minute sleep and awake-time billing.
AMD MI450X Release: 288GB HBM4, Helios Specs and Claims
#3981 November 16, 2026 AMD MI450X Release: 288GB HBM4, Helios Specs and Claims AMD adds the MI450X with 288GB of HBM4 to Helios, details the 432GB MI430X for FP64 science, and claims 21,333 tok/s per MI455X on DeepSeek-R1.
Antigravity Dynamic Workflows: How /run-workflow Works
#3977 November 16, 2026 Antigravity Dynamic Workflows: How /run-workflow Works Google Antigravity Dynamic Workflows turn a task into a reviewable Python script that runs agents in parallel or in sequence, on AI Pro and Ultra plans.
Muse Business Calls: Meta AI Calls US Firms for $20 a Month
#3968 November 16, 2026 Muse Business Calls: Meta AI Calls US Firms for $20 a Month Muse now places calls to US businesses for all paid subscribers. Plans start at $20 a month, with pricing, examples and the hang-up problem explained.
Inkling 2 Small Open Weights: 500GB VRAM, Apache 2.0
#3962 November 16, 2026 Inkling 2 Small Open Weights: 500GB VRAM, Apache 2.0 Inkling 2 Small weights arrived on Hugging Face under Apache 2.0 on November 12, 2026. NVFP4 needs 500GB VRAM and MXFP8 needs 800GB.
Edge0-145B Release: 145B On-Device Model, 29GB Peak
#3958 November 16, 2026 Edge0-145B Release: 145B On-Device Model, 29GB Peak Edge0-145B runs a 145B model on a laptop at 29GB peak memory and 80+ tokens/s. Benchmarks, architecture, license limits and unconfirmed claims explained.
Perplexity Decider v1.3 Release: Locate, Pricing and Scores
#3957 November 16, 2026 Perplexity Decider v1.3 Release: Locate, Pricing and Scores Perplexity Decider v1.3 and Flash v1.3 are open Apache decision models that return probabilities and, with locate, exact text spans. Scores and limits inside.
T3 Code UI Customization: Ask Your Agent, Watch It Change
#3994 November 15, 2026 T3 Code UI Customization: Ask Your Agent, Watch It Change T3 Code nightly builds let you ask an agent to restyle the app UI and see it update live. What shipped, how to try it, and what is still unknown.
XYBench Results: Models Mention the Fix, Miss the Goal
#3992 November 15, 2026 XYBench Results: Models Mention the Fix, Miss the Goal Ai2 XYBench tests 8,115 questions: models included the real fix in 59 to 71 percent of answers but steered users to their actual goal in only 47 to 60 percent.
ANR Neural Rendering Release: Frame Generation Guide
#3986 November 15, 2026 ANR Neural Rendering Release: Frame Generation Guide Apex ANR brings frame generation to NVIDIA, AMD and Intel GPUs in beta, with nr-record to record a scene and compare ANR FG against DLSS-G, FSR and XeSS.
Vercel UI SDK Beta: Behaviors, Release and How It Works
#3980 November 14, 2026 Vercel UI SDK Beta: Behaviors, Release and How It Works Vercel Labs UI SDK beta gives agents 50 plus behaviors instead of finished components, with adapters for five frameworks and an agent bridge.
Coding Agent Index v2.1: Scores, Cost and How It Works
#3979 November 14, 2026 Coding Agent Index v2.1: Scores, Cost and How It Works Coding Agent Index v2.1 averages 322 tasks. Fable 5.2 and Opus 5.5 tie at 58, but cost per task ranges from $1.51 to $44.22 at max effort.
Cursor Security Incident: Fireworks Key, Traces, Extension
#3978 November 14, 2026 Cursor Security Incident: Fireworks Key, Traces, Extension Cursor disclosed the full chain on November 13, 2026: an exposed Fireworks key, agent traces under 0.05 percent, and tampered Claude Code builds for up to
DeepSeek App APIs: Search, Files and Voice Pricing
#3974 November 14, 2026 DeepSeek App APIs: Search, Files and Voice Pricing DeepSeek opened its app features as APIs on November 13, 2026: web search, file input, URL reading, safety, ASR and TTS on deepseek-flash, with per-use pricing.
Codex CLI 0.169 Release: Side-Chat Tabs and --trust Gate
#3970 November 14, 2026 Codex CLI 0.169 Release: Side-Chat Tabs and --trust Gate OpenAI shipped Codex CLI 0.168 and 0.169 on November 13: side-chat tabs in a terminal workspace, task search, and a --trust gate that disables MCP in untrusted
Brave Search API x402 Payments: $0.005 Per Query, No Key
#3967 November 14, 2026 Brave Search API x402 Payments: $0.005 Per Query, No Key Brave Search API now accepts x402 agent payments: $0.005 per query in USDC on Base, no account or API key, with a 402 pay-then-retry flow and Purl Wallet.
Claude Code v2.1.323 Release: Code Mode and 1M Context
#3966 November 14, 2026 Claude Code v2.1.323 Release: Code Mode and 1M Context Claude Code v2.1.323 adds experimental Code Mode, raises the assumed context for unknown models from 200K to 1M, and gates plugins by version.
World Labs Atlas Release: Pricing, Tools and How It Works
#3965 November 14, 2026 World Labs Atlas Release: Pricing, Tools and How It Works World Labs opened Atlas to everyone on November 12: four tools, a credit-priced splat API at $0.50 to $5 per run, and no open weights.
OpenAI Daybreak Red Team and Defense Factory: Release Guide
#3960 November 14, 2026 OpenAI Daybreak Red Team and Defense Factory: Release Guide OpenAI launched Daybreak Red Team and Defense Factory on November 12, 2026: agent pen tests for web apps, plus patch PRs from GPT-6 Cyber, with no pricing yet.
Self-Hosted Claude Managed Agents: Pricing and How It Works
#3956 November 14, 2026 Self-Hosted Claude Managed Agents: Pricing and How It Works Anthropic opened self-hosted Claude Managed Agents in private beta on Nov 12, 2026. Harness, sandboxes, credentials and session history run in your own cloud.
Sierra Poppy 1.0 Release: Personal Agent Protocol Guide
#3991 November 13, 2026 Sierra Poppy 1.0 Release: Personal Agent Protocol Guide Sierra releases Poppy 1.0, the Personal Agent Protocol. Shopify enables it for every store and Meta Muse uses it by default, with OAuth control.
Google Opens Antigravity /teamwork: Proofs and Rust Gains
#3989 November 13, 2026 Google Opens Antigravity /teamwork: Proofs and Rust Gains Google opens Antigravity /teamwork to researchers and developers, reporting a 2013 conjecture proof, 800,000 lines of Rust and 4.09x kernel speedups.
Cloudflare Kitesurf Open Source: Browser Engine for Agents
#3988 November 13, 2026 Cloudflare Kitesurf Open Source: Browser Engine for Agents Cloudflare open sourced Kitesurf, its Workers only browser engine for AI agents, with CDP and Puppeteer support, Apache 2.0 and a paid plan catch.
Aion Instruct on Windows: 2.3B, 32K Context, GPU Rollout
#3987 November 13, 2026 Aion Instruct on Windows: 2.3B, 32K Context, GPU Rollout Aion Instruct replaces Phi Silica as the Windows on-device model: 2.3B parameters, a 32K context limit, GPU expansion and a January 2027 retail target.
OpenCode v2.3.0 Release: Background Subagents by Default
#3985 November 13, 2026 OpenCode v2.3.0 Release: Background Subagents by Default OpenCode v2.3.0 makes subagents run in the background by default. How immediate session IDs, idle reports, experimental MCP events and shell timeouts work.
llama.cpp v1.0.0 Release: API Stability and What Changes
#3984 November 13, 2026 llama.cpp v1.0.0 Release: API Stability and What Changes llama.cpp and ggml v1.0.0 arrived on November 13, 2026 after 3.5 years. The semantic versioning rules, 35 removed APIs, FP8 KV cache and upgrade limits.
LangChain Series C: $300M Raise, $3B Valuation and LangSmith
#3982 November 13, 2026 LangChain Series C: $300M Raise, $3B Valuation and LangSmith LangChain raised $300M in a Series C at a $3B valuation on November 13, 2026. CapitalG led the round, total funding reached $460M, and LangSmith now serves
Claude Code Projects Update: Proactive Coordinator Guide
#3976 November 13, 2026 Claude Code Projects Update: Proactive Coordinator Guide Claude Code Projects now reports progress first and answers in project chat. How threads, the coordinator, plans, and the 500-thread daily limit work.
Gemini Google Wallet Finance: Spending Insights Guide
#3975 November 13, 2026 Gemini Google Wallet Finance: Spending Insights Guide Google connected Google Wallet to Gemini on November 13, 2026. US users tag @Google Wallet to chart spending, flag subscription price rises and set budgets
ChatGPT Wallet: How It Works, Pricing and Who Can Use It
#3971 November 13, 2026 ChatGPT Wallet: How It Works, Pricing and Who Can Use It OpenAI launched ChatGPT Wallet on November 13, 2026 with Amex, Mastercard and Visa. How token payments work in dots, who can use it, fees and limits.
Claude Meetings Beta Release: Pricing and How It Works
#3964 November 13, 2026 Claude Meetings Beta Release: Pricing and How It Works Claude Meetings joins Google Meet calls as a visible participant, builds docs and whiteboards live, and writes the notes. Beta pricing and limits.
Amplitude Wave Product Agent: How It Works and Results
#3961 November 13, 2026 Amplitude Wave Product Agent: How It Works and Results Amplitude Wave uses OpenAI models to find product fixes from analytics and replays, score them, and hand coding agents a plan in 1 to 2 days.
Jensen Huang Resigns: NVIDIA CEO Transition, Kress Interim
#3955 November 13, 2026 Jensen Huang Resigns: NVIDIA CEO Transition, Kress Interim Jensen Huang resigned as NVIDIA CEO on November 12, 2026. CFO Colette Kress serves as interim CEO, Harvey Jones chairs the board, and earnings land November 17.
IBM Granite LLMON: Prompt Injection Defense, How It Works
#3954 November 13, 2026 IBM Granite LLMON: Prompt Injection Defense, How It Works Granite LLMON previews split instructions from data using special tokens. IBM reports 47-point OPI gains over the best frontier model. Limits covered.
VS Code 1.146 Release: Agent Inbox and Local Harness End
#3953 November 13, 2026 VS Code 1.146 Release: Agent Inbox and Local Harness End VS Code 1.146 adds an experimental agent Inbox, Copilot Fusion model comparison and chat windows, and sets February 2027 for Local harness removal.
Samsung Gemini Glasses Release: Pricing and How It Works
#3952 November 13, 2026 Samsung Gemini Glasses Release: Pricing and How It Works Samsung and Google Intelligent Eyewear pre-orders start at $395, launch December 10 in the US and Korea, with Gemini, a 12MP camera and 9-hour battery.
Pinky: Let AI Agents Use Your Phone, With You in Control
#3951 November 13, 2026 Pinky: Let AI Agents Use Your Phone, With You in Control Pinky lets Claude Code, Codex and other MCP agents operate a real iPhone or Android phone, while its gateway holds risky steps for your approval.
gputaskman Release: Per-Process VRAM and KV Cache View
#3950 November 13, 2026 gputaskman Release: Per-Process VRAM and KV Cache View gputaskman shows per-process VRAM, spill and LLM weights versus KV cache in a terminal UI for Windows, Linux, WSL and macOS. Install, modes and limits.
Gemini 4 Hydrogen Release: Pricing, Speed and Benchmarks
#3949 November 13, 2026 Gemini 4 Hydrogen Release: Pricing, Speed and Benchmarks Gemini 4 Hydrogen costs $1.50 input and $7.50 output per 1M tokens across a 1M context. Pricing, 57% Terminal-Bench, Orchestrator and limits.
OpenSpec 2.0 Release: Shared Skills, No /opsx Commands
#3947 November 13, 2026 OpenSpec 2.0 Release: Shared Skills, No /opsx Commands OpenSpec 2.0 moves skills into one shared .agents/skills folder, removes every /opsx:* command, and drops Amazon Q. Upgrade steps and risks inside.
Google Flow: Five New Creator Tools, Character to Final Cut
#3959 November 12, 2026 Google Flow: Five New Creator Tools, Character to Final Cut Google released five creator-built tools in Flow on November 12, 2026, covering character consistency, retro styles, light effects and phone footage polish.
Seedance 2.5 flash Release: 2x Speed, Pricing and Limits
#3948 November 12, 2026 Seedance 2.5 flash Release: 2x Speed, Pricing and Limits Seedance 2.5 flash doubles generation speed and cuts 720P pricing about 34% to 0.99 yuan per second. Specs, rollout dates, limits and open questions.
Grok Bot 1.0 Release: Team Sharing, Meetings and Pricing
#3944 November 12, 2026 Grok Bot 1.0 Release: Team Sharing, Meetings and Pricing SpaceXAI released Grok Bot 1.0 on November 11, 2026. Shared bots join Slack, Microsoft Teams and video calls, propose work first and learn from the team.
Runway Praxis-1 Open Weights: 46.2% on RoboLab-120
#3941 November 12, 2026 Runway Praxis-1 Open Weights: 46.2% on RoboLab-120 Runway released Praxis-1 open weights under Apache 2.0 on November 11, 2026. The 27B model scored 46.2% on RoboLab-120 and ships with fine-tuning code.
Anthropic Mythos AMD Port: Claude on MI355X in 3 Days
#3933 November 12, 2026 Anthropic Mythos AMD Port: Claude on MI355X in 3 Days Anthropic says Mythos ported Claude prompt processing to AMD MI355X GPUs in about 3 days, writing nearly 100% of kernels, versus 6 to 12 months traditionally.
Gemini Decisions API: Pricing, Speed and How It Works
#3932 November 12, 2026 Gemini Decisions API: Pricing, Speed and How It Works Google Gemini Decisions API scores up to 1,000 questions in one pass at 150 to 250ms, priced at $0.10 per million input tokens with no output charge.
AWS Emulator Release: Local Testing, Pricing and Limits
#3931 November 12, 2026 AWS Emulator Release: Local Testing, Pricing and Limits AWS Emulator runs AWS APIs locally at 127.0.0.1:9297 with no extra service charge. The Rust tool was built entirely by autonomous agents and released under CC0.
Claude Cage Release: Isolated VM, 3,181 Stars in 11 Days
#3930 November 12, 2026 Claude Cage Release: Isolated VM, 3,181 Stars in 11 Days Claude Cage is a lightweight VM built only for Claude. It isolates network and fingerprint settings from the host and reached 3,181 stars in 11 days.
Jane Street Agent Skills Release: 5,794 Stars in 8 Days
#3929 November 12, 2026 Jane Street Agent Skills Release: 5,794 Stars in 8 Days Jane Street published agent_skills on GitHub under MIT. Its only skill, socratic-tutor, reached 5,794 stars and 401 forks in about eight days.
Gemini Intelligence on Pixel 10: Rambler and Automation
#3926 November 12, 2026 Gemini Intelligence on Pixel 10: Rambler and Automation Google expands Gemini Intelligence to Pixel 10, adds Rambler and wider task automation, and brings the agent suite to a first Motorola phone.
Plurai Simulation Release: Agent Digital Twins, $13M Seed
#3927 November 11, 2026 Plurai Simulation Release: Agent Digital Twins, $13M Seed Plurai Simulation builds a digital twin of an AI agent and its environment, surfaces failures before deployment, and post-trains fixes. Free to start, $13M
ERNIE 5.5 Flash Preview: Release, Pricing and Game Demo
#3925 November 11, 2026 ERNIE 5.5 Flash Preview: Release, Pricing and Game Demo Baidu ERNIE 5.5 Flash Preview activates 14B of 280B parameters, offers a 1M context on Qianfan, and built a full Unreal Engine 5.8 game in one 11-hour run.
BotPsy 0.8B Release: On-Device Agent Scores and Limits
#3924 November 11, 2026 BotPsy 0.8B Release: On-Device Agent Scores and Limits BotPsy is a 0.8B on-device agent model from Tether AI Research. It beats a 2.5x larger model on five of six tool benchmarks, with open Apache 2.0 weights.
Runpod for Agents Release: MCP, 35 Skills, Pricing and Setup
#3921 November 11, 2026 Runpod for Agents Release: MCP, 35 Skills, Pricing and Setup Runpod for Agents adds a hosted MCP server and 35 verified skills, 10 core and 25 specialized, so coding agents can run GPU infrastructure work.
Git 2.56.1 Security Release: Six Vulnerabilities Fixed
#3920 November 11, 2026 Git 2.56.1 Security Release: Six Vulnerabilities Fixed Git 2.56.1 fixes six vulnerabilities, including CVE-2026-111760, where a recursive clone can run attacker code. Backports reach 2.50.2.
Qwen4-Max-Preview Release: 1M Context, No Benchmarks Yet
#3919 November 11, 2026 Qwen4-Max-Preview Release: 1M Context, No Benchmarks Yet Qwen4-Max-Preview ships with a 1M token context, 393,216 token max output and reasoning controls, but Alibaba has published no benchmarks for it yet.
Gemini Robotics 2.5 Release: 96% on a Mercedes-Benz Line
#3916 November 11, 2026 Gemini Robotics 2.5 Release: 96% on a Mercedes-Benz Line Google DeepMind released Gemini Robotics 2.5 on November 10, 2026. Apollo 2 hit 96% on Mercedes-Benz kitting, with safety gains and a new jailbreak benchmark.
MiniMax H3-Image Open Weights: 16MP, 6 Steps, License Limits
#3910 November 11, 2026 MiniMax H3-Image Open Weights: 16MP, 6 Steps, License Limits MiniMax released H3-Image open weights on November 10, 2026: 16 megapixels in 6 steps, but Korea, the US, EU and UK are excluded from the license.
Expo Exact: Write Once, Run Natively on Four Platforms
#3908 November 11, 2026 Expo Exact: Write Once, Run Natively on Four Platforms Expo Exact writes an app once in Contract and runs it natively on web, macOS, iOS and Linux, with 100ms budgets and agent testing on every host.
Claude Sheets, Whiteboards and Data: Release and Pricing
#3907 November 11, 2026 Claude Sheets, Whiteboards and Data: Release and Pricing Anthropic launched Claude Sheets and Whiteboards in beta and Claude Data in GA on November 10, 2026, for all paid plans, with Meetings teased for Google Meet.
Instinct Coder 355: On-Prem AI Coding, 85 Devs Per Node
#3906 November 11, 2026 Instinct Coder 355: On-Prem AI Coding, 85 Devs Per Node AMD, Dell and Spectro Cloud launched Instinct Coder 355 on November 10, 2026: eight MI355X GPUs, 3 TB memory and GLM-5.3 support up to 85 developers per node.
Gemini 4 Carbon Release: Pricing, Access and How It Works
#3905 November 11, 2026 Gemini 4 Carbon Release: Pricing, Access and How It Works Google opened Gemini 4 Carbon to paid Gemini API developers on November 10, 2026, at $2 input and $10 output per million tokens until November 30.
AlphaProof Code: Signal Proof, 6.5M Lean Lines in 4 Weeks
#3901 November 11, 2026 AlphaProof Code: Signal Proof, 6.5M Lean Lines in 4 Weeks DeepMind's AlphaProof Code proved Signal's compiled libsignal code in Lean: about 6.5 million proof lines in 4 weeks, with clear scope limits.
browser-use 1.0 Release: CDP Direct, 80% Code Deleted
#3900 November 10, 2026 browser-use 1.0 Release: CDP Direct, 80% Code Deleted browser-use 1.0.0 deletes about 136,866 lines, lets the agent drive Chrome directly over CDP from a Python REPL, and changes the default model, browser and run
NVIDIA Invests in Four Korean Startups: Point2, Upstage
#3899 November 10, 2026 NVIDIA Invests in Four Korean Startups: Point2, Upstage NVIDIA announced direct investments in Point2, Upstage, Motif and Realworld on November 9, 2026, plus an NVLink Fusion MOU linking Rebellions NPUs to its racks.
Seedream 5.5 Release: 100-Round Editing, Pricing and Scores
#3895 November 10, 2026 Seedream 5.5 Release: 100-Round Editing, Pricing and Scores ByteDance Seed released Seedream 5.5 on November 9, 2026. It edits one image for 100 rounds, generates native RGBA assets and cuts severe text errors to 0.4%.
NVIDIA GR00T 2 Release: Performance, Data and How It Works
#3894 November 10, 2026 NVIDIA GR00T 2 Release: Performance, Data and How It Works NVIDIA unveiled GR00T 2 at CoRL on November 9, 2026. Trained on 402,730 hours of human video, it hit 40% zero-shot and 95% with five demos on unseen tasks.
Cloudflare Codex Harness: Durable Objects, Resume Guide
#3892 November 10, 2026 Cloudflare Codex Harness: Durable Objects, Resume Guide Cloudflare put the Codex agent loop inside a Durable Object on November 9, 2026. Tasks, Sessions, Streams and Workspace let interrupted coding agents resume.
Claude Fable 5.2 and Mythos 5.2: Pricing and Performance
#3889 November 10, 2026 Claude Fable 5.2 and Mythos 5.2: Pricing and Performance Claude Fable 5.2 costs $8/$40 per million tokens, runs up to 30% faster and scores 49.1% on FrontierCode 2.0. Mythos 5.2, pricing and limits explained.
Meta Muse Ember Release: Agent Pricing and Performance
#3888 November 10, 2026 Meta Muse Ember Release: Agent Pricing and Performance Meta Muse Ember costs $0.10 per million input tokens, matches Haiku at 36 on the Intelligence Index and scores 46 on the Coding Agent Index.
Spacefast Release: Deploy Coding Agent Apps, Pricing
#3887 November 10, 2026 Spacefast Release: Deploy Coding Agent Apps, Pricing Spacefast from Automattic lets Claude Code, Codex and Cursor apps go live on private links. Pricing, supported stacks and the lock-in question, explained.
Google TPU 8tx and 8i Max Release: Confidential, Big Context
#3886 November 10, 2026 Google TPU 8tx and 8i Max Release: Confidential, Big Context Google adds TPU 8tx for confidential inference and TPU 8i Max with two 8i chips to the eighth TPU generation, plus TPU Sync and GKE dynamic slicing GA.
phoneuse: Control an Unmodified iPhone From a Linux PC
#3881 November 9, 2026 phoneuse: Control an Unmodified iPhone From a Linux PC phoneuse lets agents see and control an unmodified iPhone from a Linux PC: AirPlay and UxPlay supply the screen, Bluetooth HID supplies taps and typing.
Open WebUI v0.13.1 Release: Coding Agents and Tether
#3878 November 9, 2026 Open WebUI v0.13.1 Release: Coding Agents and Tether Open WebUI v0.13.1 puts Claude Code, Codex and Pi in the model picker, adds Tether access to your own computer, an Auto tool mode and a security fix.
Meta Muse Spaces: Custom App Spaces You Describe in Words
#3868 November 9, 2026 Meta Muse Spaces: Custom App Spaces You Describe in Words Meta launched Spaces in Muse on November 6, 2026. Describe an idea and Muse builds a custom space in the app, from gift trackers to travel planners.
Aaronson-Ambainis Proof: AI System Proves 2008 Conjecture
#3863 November 9, 2026 Aaronson-Ambainis Proof: AI System Proves 2008 Conjecture Physical Superintelligence says its system, running an undisclosed Anthropic model, proved the Aaronson-Ambainis conjecture open since 2008, with Lean
CompressionBench Results: Coding Agents Compress LLMs
#3831 November 9, 2026 CompressionBench Results: Coding Agents Compress LLMs CompressionBench tests whether coding agents can compress LLMs alone. 29 of 36 submissions passed verification, with accuracy at 77.33% vs 79.56% for BF16.
LaunchDarkly Release Autopilot: AI Code Release Control
#3817 November 9, 2026 LaunchDarkly Release Autopilot: AI Code Release Control LaunchDarkly Release Autopilot scores pull request risk, adds feature flags, rolls out in stages and rolls back on guardrail metrics. GA is November 16.
NVIDIA Reshard Release: 2.7x to 5.7x Faster RL Refit
#3877 November 8, 2026 NVIDIA Reshard Release: 2.7x to 5.7x Faster RL Refit NVIDIA Reshard moves RL weights DTensor to DTensor, sending only overlapping shards. NeMo RL refits ran 2.7x to 5.7x faster in NVIDIA tests.
Microsoft ThinkingBox Results: RL Training and Reliability
#3876 November 8, 2026 Microsoft ThinkingBox Results: RL Training and Reliability Microsoft ThinkingBox graded agents by database state, not final answers. RL training lifted all-pass tasks from 38 to 107 and pass@1 to 60.78%.
Google ADK v2.14.0: Stable Environments, A2A and Evals
#3875 November 8, 2026 Google ADK v2.14.0: Stable Environments, A2A and Evals Google released ADK Python v2.14.0 on November 7, 2026. Environments, A2A and evaluation are now stable, with Teams support, evals and breaking changes.
AMD PrimusClaw Release: Open-Source Managed Agent Runtime
#3871 November 8, 2026 AMD PrimusClaw Release: Open-Source Managed Agent Runtime AMD open-sourced PrimusClaw, a managed agent runtime for your own Kubernetes cluster beside AMD Instinct GPUs, with leased runs that survive failures.
Meta Muse Wonder Specialized Models: Memory and Focus Gains
#3870 November 8, 2026 Meta Muse Wonder Specialized Models: Memory and Focus Gains Meta began rolling out specialized models trained with Muse Wonder to Muse and Meta AI on November 6, 2026, adding Active Memory and Active Focus.
JetBrains Air in 2026.3: ACP Agents Inside Every IDE
#3828 November 8, 2026 JetBrains Air in 2026.3: ACP Agents Inside Every IDE JetBrains shipped Air in the 2026.3 IDE lineup on November 5, 2026, running ACP agents inside the IDE, with Junie Lite included free.
METR: Anthropic Model Improvement Is Running 2x Faster
#3827 November 8, 2026 METR: Anthropic Model Improvement Is Running 2x Faster METR high-access review finds Anthropic improving model capability about twice as fast since October 2025, with AECI growth at 1.9x and wide uncertainty.
LeRobot v0.7.0 Release: Prompt Robots, Stream 300 TB
#3860 November 7, 2026 LeRobot v0.7.0 Release: Prompt Robots, Stream 300 TB Hugging Face LeRobot v0.7.0 lets you prompt a running robot, adds eight world and vision-language models, and streams 300 TB of data without download.
Hermes Agent v0.23.0: One Gateway, Browser Workspace
#3851 November 7, 2026 Hermes Agent v0.23.0: One Gateway, Browser Workspace NousResearch Hermes Agent v0.23.0 unifies every client on one gateway, moves the desktop workspace into a browser tab, and adds priced MCP tools.
Claude Code 2.1.319: MCP Tasks Go Background, Release Notes
#3850 November 7, 2026 Claude Code 2.1.319: MCP Tasks Go Background, Release Notes Claude Code 2.1.319 sends MCP tasks to the background by default, adds denyInsteadOfPrompting, and changes auto mode checks behind LLM gateways.
Kiro Bots Release: Persistent Agents, Pricing and Memory
#3849 November 7, 2026 Kiro Bots Release: Persistent Agents, Pricing and Memory AWS Kiro Bots are generally available in Kiro Crew: persistent role based agents with per bot memory, scoped permissions and credit based pricing.
Mistral Studio Managed Agents and Apps: Release Guide
#3844 November 7, 2026 Mistral Studio Managed Agents and Apps: Release Guide Mistral Studio adds Managed Agents and Apps on November 5, with cloud, VPC and on-premises deployment, OpenTelemetry tracing and pre-release evaluation.
Cursor Live Apps: Release, Postgres, Auth and Deployment
#3841 November 7, 2026 Cursor Live Apps: Release, Postgres, Auth and Deployment Cursor Live Apps builds a web app in the editor, adds Postgres and Cursor SSO, provisions hosting and returns a live link after checks.
AI Frontiers Is Over: AutoGen Archived After Microsoft Move
#3874 November 6, 2026 AI Frontiers Is Over: AutoGen Archived After Microsoft Move John Langford announced the end of Microsoft Research AI Frontiers on November 6, 2026. AutoGen is archived, Fara is in limbo, and teams must plan a move.
HyperFlow H3 Flash Release: Performance, Size and License
#3873 November 6, 2026 HyperFlow H3 Flash Release: Performance, Size and License HyperFlow H3 Flash shrinks the MiniMax-H3 transformer to 13.7B, generates video with stereo audio in 4 to 8 steps, and runs in a 9.1 GB GGUF file.
Cursor Voice Chat with Agents: Real-Time Coding Calls
#3872 November 6, 2026 Cursor Voice Chat with Agents: Real-Time Coding Calls Cursor launched Voice Chat with Agents on November 6, 2026: talk to a coding agent in real time on desktop, web and iOS while it works, powered by Grok Voice.
GitHub Copilot in Slack and Teams: Generally Available
#3867 November 6, 2026 GitHub Copilot in Slack and Teams: Generally Available GitHub Copilot in Slack and Teams is generally available: investigate bugs, create issues and open pull requests from chat, with Canvas review in Microsoft
Claude Code Cloud Auto Setup: Plans and How It Works
#3866 November 6, 2026 Claude Code Cloud Auto Setup: Plans and How It Works Claude Code cloud sessions now set up their own environment. Claude detects dependencies, network access and API keys, and saves the setup for the next session.
Copilot Code Review API Management: Settings and Models
#3865 November 6, 2026 Copilot Code Review API Management: Settings and Models GitHub now exposes Copilot code review settings through REST and GraphQL APIs, and balanced reviews split work across three reviewer models plus a merger.
Anthropic Agent Memory: Dreaming, Wiki Scores and Design
#3864 November 6, 2026 Anthropic Agent Memory: Dreaming, Wiki Scores and Design Anthropic maps agent memory: dreaming fixes notes offline, wiki scores rise from 13% to 48%, and small curated files beat hundreds of scattered ones.
DeepMind RSA-1024 Attacks: 1,600x Forgery, 30x Factoring
#3861 November 6, 2026 DeepMind RSA-1024 Attacks: 1,600x Forgery, 30x Factoring Google DeepMind's human-supervised AI agents produced two faster attacks on legacy RSA-1024: a 1,600x theoretical forgery gain and 30x cheaper batch factoring.
Salesforce Koa Decisions: Gemma Decision Model Release
#3848 November 6, 2026 Salesforce Koa Decisions: Gemma Decision Model Release Salesforce Koa Decisions, built on Gemma, adds image and audio inputs and returns a choice, a yes or no, or a score with a confidence value in milliseconds.
Ai2 ToyBlocks Release: Robot Learning Testbed, 85,336 Demos
#3847 November 6, 2026 Ai2 ToyBlocks Release: Robot Learning Testbed, 85,336 Demos Ai2 released ToyBlocks on November 6, 2026, an open robot learning testbed with 85,336 auto generated demonstrations, 144 million frames and two trained models.
Antigravity Remote Agents: Cloud Coding Sessions, Pricing
#3846 November 6, 2026 Antigravity Remote Agents: Cloud Coding Sessions, Pricing Google Antigravity Remote Agents run coding sessions in an isolated Google-managed cloud environment, so work continues after you close your laptop.
Alpie Seed 9B Release: Offline Coding, 61.2 SWE-bench
#3845 November 6, 2026 Alpie Seed 9B Release: Offline Coding, 61.2 SWE-bench 169Pi released Alpie Seed 9B open weights on November 6, 2026. The 9B vision language model codes offline on laptops and phones under Apache 2.0.
Dropbox Tigris Acquisition: $350M DBX Agent Cloud Launch
#3840 November 6, 2026 Dropbox Tigris Acquisition: $350M DBX Agent Cloud Launch Dropbox agreed to buy Tigris Data for $350 million in cash, its largest deal ever, and launched DBX Agent Cloud on the Tigris storage platform.
Meta Muse Voice Release: Realtime Avatar Video Calls
#3839 November 6, 2026 Meta Muse Voice Release: Realtime Avatar Video Calls Meta opened Muse Voice to all US users on November 5, 2026, adding phone and video calls with its Muse agent and a realtime avatar.
NVIDIA Vera CPU Agent Performance: 1.6x Throughput
#3838 November 6, 2026 NVIDIA Vera CPU Agent Performance: 1.6x Throughput NVIDIA Vera CPU delivered 1.6x more agents per watt and 2x agent speed versus x86 in a 100 MW test, with six partners reporting production results.
Kakao Kanana-2-v-decide Release: 0.04s Decision Model
#3837 November 6, 2026 Kakao Kanana-2-v-decide Release: 0.04s Decision Model Kakao released Kanana-2-v-decide as open weights on November 6, 2026. The 3.67B multimodal model answers with probabilities in under 0.04 seconds on H200.
GPT-6 Cyber Release: ExploitGym 48.5% and Critical Rating
#3825 November 6, 2026 GPT-6 Cyber Release: ExploitGym 48.5% and Critical Rating OpenAI released GPT-6 Cyber on November 5, 2026 for approved Daybreak Red customers. It scores 48.5% on ExploitGym and carries a Critical cyber rating.
rea-video Release: Shot-by-Shot Video Reverse Engineering
#3824 November 6, 2026 rea-video Release: Shot-by-Shot Video Reverse Engineering rea-video reverse engineers any video shot by shot, classifying each shot as filmed, AI generated or coded, then outputs a remake plan and report.html.
Perplexity Teammates Release: Persistent Agents, Pricing
#3823 November 6, 2026 Perplexity Teammates Release: Persistent Agents, Pricing Perplexity launched Teammates on November 5, 2026: named agents on Computer with standing briefs, separate sessions and Max pricing at $200 a month.
Baidu ERNIE-5.5-Flash-Base: 280B Release, License Limits
#3822 November 6, 2026 Baidu ERNIE-5.5-Flash-Base: 280B Release, License Limits Baidu released ERNIE-5.5-Flash-Base on Nov 5, 2026: 280B total, 14B active, 93.3 MMLU, but its license excludes South Korea, the EU and the UK.
PLaMo 3 Translate Open Weights: 0.8B to 31B, 53 Languages
#3821 November 6, 2026 PLaMo 3 Translate Open Weights: 0.8B to 31B, 53 Languages PFN released PLaMo 3 Translate weights in four sizes from 0.8B to 31B, covering 53 languages, offline phone use and a revenue based free licence.
DigitalOcean Cuttlefish Release: iPhone Agent Supervision
#3820 November 6, 2026 DigitalOcean Cuttlefish Release: iPhone Agent Supervision DigitalOcean launched Cuttlefish on November 5, 2026, an iPhone app for Managed Agents that streams sessions live and gates tool calls with Face ID.
Gemini Omni 1.2 Flash GA: 15-Second Video, $0.07 Pricing
#3819 November 6, 2026 Gemini Omni 1.2 Flash GA: 15-Second Video, $0.07 Pricing Google made Gemini Omni 1.2 Flash generally available with native 15-second video, 720p pricing at $0.07 per second and conversational editing in one clip.
Kimi Hosted Agents: Pricing, Skills and How It Works
#3818 November 6, 2026 Kimi Hosted Agents: Pricing, Skills and How It Works Moonshot AI has formally launched Kimi Hosted Agents, a managed agent runtime with 140+ skills, 30+ plugins, 12 templates and sandbox time at 0.072 yuan per
Inkling 2 Small Release: 17B Active, Pricing and Scores
#3842 November 5, 2026 Inkling 2 Small Release: 17B Active, Pricing and Scores Thinking Machines Inkling 2 Small activates 17B of 594B parameters, scores 31 on the Intelligence Index, and heads to Apache 2.0 weights next week.
Claude Code Lead Cat Wu Joins OpenAI Core Products
#3816 November 5, 2026 Claude Code Lead Cat Wu Joins OpenAI Core Products Claude Code product lead Cat Wu joined OpenAI as Core Products PM lead on November 5, reporting to Tibo Sottiaux. Anthropic has not named a successor.
ToolSpace: MCP Agent Training From Real Execution Traces
#3800 November 5, 2026 ToolSpace: MCP Agent Training From Real Execution Traces ToolSpace builds MCP agent training environments from real execution traces. Qwen3.6 rose from 23.0 to 41.0 and GLM-5.2 reached 56.6, on par with Claude Opus 5.
Cognition SWE-2 Flash: Pricing, Speed and Benchmarks
#3795 November 5, 2026 Cognition SWE-2 Flash: Pricing, Speed and Benchmarks Cognition SWE-2 Flash costs one fifth of SWE-2, runs 3x faster and is free until November 30. Its benchmark gains and closed weights, explained.
Ricursive Ripple-65B: Open-Weight Coding Model Benchmarks
#3794 November 5, 2026 Ricursive Ripple-65B: Open-Weight Coding Model Benchmarks Ricursive released Ripple-65B Preview, a 65B dense open-weight coding model that reports 80.5 on Terminal-Bench 2.1, ahead of the 501B Beam. All scores are
Kimi K3.1 Pro Open Weights: 2.8T MoE Release Explained
#3793 November 5, 2026 Kimi K3.1 Pro Open Weights: 2.8T MoE Release Explained Moonshot AI published the full Kimi K3.1 Pro weights on Hugging Face: a 2.8T MoE model with 104B active parameters, a 1M token context and MXFP4 weights.
Nooks Node and Revenue Agent Platform: $500M Growth Deal
#3791 November 5, 2026 Nooks Node and Revenue Agent Platform: $500M Growth Deal Nooks secured a $500M growth commitment from General Catalyst and launched Node, an AI sales assistant, plus its Revenue Agent Platform for 2,000+ customers.
Enterprise AI Spending Hits $176.8B: Menlo Ventures Report
#3790 November 5, 2026 Enterprise AI Spending Hits $176.8B: Menlo Ventures Report Menlo Ventures put enterprise AI deployment spending at $176.8 billion in 2026, up 5.4x in a year, with agents taking 80 percent of OpenRouter tokens.
OpenAI Acquires Pierre Computer: Code Storage for Codex
#3788 November 5, 2026 OpenAI Acquires Pierre Computer: Code Storage for Codex OpenAI agreed on November 4, 2026 to acquire Pierre Computer, maker of Trees, Diffs and Code Storage, and fold its machine-first Git storage into Codex.
agent-autoloop: Background CLI Agents With Decision Cards
#3787 November 5, 2026 agent-autoloop: Background CLI Agents With Decision Cards agent-autoloop runs CLI coding agents in the background on macOS and Linux, pausing only to send a decision card when human judgment is needed.
Tencent Hy4 Release: 770B MoE With Vision and Computer Use
#3784 November 5, 2026 Tencent Hy4 Release: 770B MoE With Vision and Computer Use Tencent's Hy4 pairs a 770B MoE with first-time vision and Computer Use, cuts self-measured hallucination to 5.45%, and still ships without public weights.
Claude Animated Shorts: Code-Rendered Videos with Remotion
#3724 November 5, 2026 Claude Animated Shorts: Code-Rendered Videos with Remotion A Claude skill renders 55 to 70 second animated shorts locally with three.js and Remotion, with no AI video generator and a full example included.
Claude Code Background Subagents: Status Cost Cut by 99%
#3814 November 4, 2026 Claude Code Background Subagents: Status Cost Cut by 99% Anthropic moved Claude Code background subagent status lines to Haiku 5.5 writing from a short note, cutting the per-update cost by about 99 percent.
NVIDIA Attribution: Root Cause for Distributed AI Failures
#3781 November 4, 2026 NVIDIA Attribution: Root Cause for Distributed AI Failures NVIDIA open sourced Attribution, its rule-based system for diagnosing distributed AI workload failures, run internally for years on millions of weekly
Sarvam Prism: Open Decision Model, 44ms, 71.8 Average
#3780 November 4, 2026 Sarvam Prism: Open Decision Model, 44ms, 71.8 Average Sarvam released Prism, an open-weight LoRA decision model on Sarvam 105B, scoring 71.8 at 44ms with calibrated choice probabilities under Apache 2.0.
Hallmark vs Impeccable: Which Design Skill Fits the Job
#3776 November 4, 2026 Hallmark vs Impeccable: Which Design Skill Fits the Job Independent tests compare Hallmark and Impeccable design skills. Hallmark led on new landing pages, Impeccable on fixing existing screens and accessibility.
BLA Harden: How It Blocks AI Reverse Engineering Agents
#3746 November 4, 2026 BLA Harden: How It Blocks AI Reverse Engineering Agents BLA hardens binaries, Electron apps, JS bundles, Python and WebAssembly against AI reverse engineering agents with Poison, Exhaust and Hallucinate layers.
AnyModelBot: Local-First AI Bot With 70+ Models, No Lock-In
#3745 November 4, 2026 AnyModelBot: Local-First AI Bot With 70+ Models, No Lock-In AnyModelBot is an open source, local-first AI bot that connects GPT, Claude, Gemini, Grok and 70+ models through one Modelflare key, with per-bot memory.
Latent Sparse Attention: 72.4x Prefill at 10M Tokens
#3742 November 4, 2026 Latent Sparse Attention: 72.4x Prefill at 10M Tokens Prefrontal AI released Latent Sparse Attention as open research: 72.4x prefill at 10M tokens on H200, with accuracy held within 2 points in its own tests.
OpenAI LifeVal: GPT-6 Astra Scores 80.6% on Daily Tasks
#3740 November 4, 2026 OpenAI LifeVal: GPT-6 Astra Scores 80.6% on Daily Tasks OpenAI released LifeVal on November 2, 2026, a benchmark for everyday tasks. GPT-6 Astra averaged 80.6% reward, but task count and scoring stay undisclosed.
Verlintas NOIR: Android Privacy Audit via adb, 1,187 Stars
#3723 November 4, 2026 Verlintas NOIR: Android Privacy Audit via adb, 1,187 Stars Verlintas NOIR audits Android apps on a desktop via adb, mapping permissions, network endpoints and risk scores. It reached 1,187 stars in two days.
AiSOC: Open-Source AI SOC With Full Agent Audit Trails
#3719 November 4, 2026 AiSOC: Open-Source AI SOC With Full Agent Audit Trails AiSOC is a self-hosted open-source AI SOC that logs every agent prompt and tool call, runs 2,538 active rules and gates external actions on human approval.
stack-diagram: One Sentence to an Architecture Diagram
#3718 November 4, 2026 stack-diagram: One Sentence to an Architecture Diagram stack-diagram is an agent skill by caru-ini that turns one sentence about a stack into an architecture diagram with official logos, as a single offline HTML
StateBench: Frontier Models Fall to 13% on State Tracking
#3810 November 3, 2026 StateBench: Frontier Models Fall to 13% on State Tracking Ai2 StateBench tests four tiers of state tracking over long documents. Frontier models fall to 13% or less on the hardest tier, while small recurrent models
OpenAI Bug Bounty: Critical Payouts Rise to $172,000
#3736 November 3, 2026 OpenAI Bug Bounty: Critical Payouts Rise to $172,000 OpenAI raised Bugcrowd bounty caps on November 2. Critical findings in core products pay up to $30,000, and exceptional findings reach $172,000 with bonuses.
Anthropic S-1/A: $16.96B Revenue and the RSI Warning
#3730 November 3, 2026 Anthropic S-1/A: $16.96B Revenue and the RSI Warning Anthropic's amended S-1/A adds $16.96B Q3 revenue, $895.4B in committed compute spend, and a warning that it may imminently meet its RSI threshold.
Weight-Space Programming: LLM Edits Weights, MATH-500 Gains
#3728 November 3, 2026 Weight-Space Programming: LLM Edits Weights, MATH-500 Gains MIT CSAIL researchers report an LLM editing raw neural network weights with no optimizer. Changing 0.016% of Qwen3.5-0.8B lifted MATH-500 from 7.9% to 48.1%.
Floorplan Maps Your Codebase for Claude Code and Codex
#3726 November 3, 2026 Floorplan Maps Your Codebase for Claude Code and Codex Floorplan turns a codebase read by Claude Code or Codex into an interactive app map with health checks for exposed keys, open tables and unauthenticated
Hallmark v2 Design Skill: 50 Themes for Coding Agents
#3725 November 3, 2026 Hallmark v2 Design Skill: 50 Themes for Coding Agents Hallmark v2 topped GitHub weekly trending with 11,600 new stars. It adds 50 themes, redesign and audit commands, and checks pages at eight widths.
OpenIDEA Rebuilds IntelliJ IDEA in Rust: 169 MB Memory
#3710 November 3, 2026 OpenIDEA Rebuilds IntelliJ IDEA in Rust: 169 MB Memory OpenIDEA rebuilds the IntelliJ IDEA workflow in Rust: 169 MB memory against 2,011 MB, ready in 1.1 s against 13.2 s in README benchmarks, with an
ServiceNow in ChatGPT: OpenAI MCP Plugin, Action Fabric
#3639 November 3, 2026 ServiceNow in ChatGPT: OpenAI MCP Plugin, Action Fabric ServiceNow and OpenAI expanded their enterprise AI partnership on October 28, 2026. An MCP plugin runs ServiceNow workflows inside ChatGPT, controlled via
MiniMax mcode-stack: 17 Open-Source Agent Skills and Setup
#3615 November 3, 2026 MiniMax mcode-stack: 17 Open-Source Agent Skills and Setup MiniMax open-sourced mcode-stack on October 28, 2026: 17 agent skills its Code team uses daily, in SKILL.md format that also installs into Claude Code.
Coding Agents Made CI the Bottleneck: Anthropic and Linear
#3557 November 3, 2026 Coding Agents Made CI the Bottleneck: Anthropic and Linear Coding agents made CI the bottleneck: Anthropic's CI load grew 25x in six months, daily merges from 2,500 to 30,000. Headline figures are secondhand quotes.
day-trader-skill: 11 Claude Skills, Time Machine Replays
#3542 November 3, 2026 day-trader-skill: 11 Claude Skills, Time Machine Replays day-trader-skill packs 11 Claude skills for day trading and drew 2,052 GitHub stars in 7 days. A Time Machine replays 10,000 outcomes; paper trading is the
gpu-unlock for RTX 30: BF16 Gains up to 80% Claimed
#3529 November 3, 2026 gpu-unlock for RTX 30: BF16 Gains up to 80% Claimed Zagitta's gpu-unlock lifts the FP/BF16 tensor core rate limit on RTX 30 cards. On an RTX 3090 Ti it reports BF16 up 80% and TF32 up 94%.
Claude Security: Continuous Monitoring Rescans Every Commit
#3484 November 3, 2026 Claude Security: Continuous Monitoring Rescans Every Commit Anthropic shifted Claude Security to continuous monitoring: connect a repository and every code change triggers an automatic rescan, with confirmed fixes
NVIDIA GTC Berlin Open Models: Six Families, Open Weights
#3386 November 3, 2026 NVIDIA GTC Berlin Open Models: Six Families, Open Weights NVIDIA announced six open model families at GTC Berlin: language, robotics, biology, and quantum, including a RAG retriever and a robotics world model as open
DigUp: On-Device Semantic File Search, EmbeddingGemma 2
#3382 November 3, 2026 DigUp: On-Device Semantic File Search, EmbeddingGemma 2 DigUp is a cross-platform, open-source semantic file search tool. It embeds photos, PDFs, recordings, and videos with EmbeddingGemma 2 and keeps all data on
ElevenLabs Eleven Realtime v1: Speech-to-Speech Voice AI
#3379 November 3, 2026 ElevenLabs Eleven Realtime v1: Speech-to-Speech Voice AI ElevenLabs' Eleven Realtime claims 98.8% on Big Bench Audio and sub-one-second first response for speech-to-speech voice agents, with an OpenAI-compatible API.
TorchTPU: Native PyTorch Backend for TPUs, 52% MFU Claimed
#3377 November 3, 2026 TorchTPU: Native PyTorch Backend for TPUs, 52% MFU Claimed TorchTPU is Google and Meta's open-source PyTorch backend for TPUs. Existing PyTorch code moves over with a device change and a few lines, with 52% MFU claimed.
huashu-motion: Coding-Agent Skill That Makes Launch Videos
#3339 November 3, 2026 huashu-motion: Coding-Agent Skill That Makes Launch Videos huashu-motion is a coding agent skill that turns one prompt into a launch video: research, composition, editing, rendering. MIT-licensed, 200 stars in about
GPT-6 Pro Parallel Agents: One Request, Multiple Agents
#3330 November 3, 2026 GPT-6 Pro Parallel Agents: One Request, Multiple Agents OpenAI's October 16 release notes introduced GPT-6 Pro Parallel Agents: one request fans out to multiple agents, with per-agent progress and individual result
FlashProve: Provable Inference for LLMs, Open Source
#3748 November 2, 2026 FlashProve: Provable Inference for LLMs, Open Source Anthropic open sourced FlashProve, a prototype that proves which model produced an answer without revealing weights, at 110 tokens per second on one H100.