Anthropic swaps git for in-house Scion in five months
Anthropic says it moved its entire monorepo from git to a homegrown system called Scion in five months, after daily merges grew more than tenfold, with Claude doing much of the work.
AI Coding News and Analysis
165 verified stories covering AI Coding, product updates and industry developments. Page 1 of 2
Epic goes AI-first, delaying its Disney shooter
Bloomberg reports Epic Games is shifting development to “AI first” and has delayed its Disney shooter indefinitely, with Claude writing the code. Tim Sweeney says it isn't about layoffs.
Meta Launches Muse Wonder as API Output Price Rises Nearly 5x
Meta's Muse Wonder lifts the Artificial Analysis index to 52, but output tokens now cost $20 per million, up from $4.25. The stronger Full tier is Muse Code only.
QGIS Votes to Ban AI in Code Contributions
QGIS has merged QEP 428, "No AI contributions permitted," replacing its looser human-in-the-loop policy, though enforcement and non-code content remain unclear.
Claude-built Adobe clones now target Microsoft Office
Brandon Thomas's Rust-based ArtCraft series has grown from seven Adobe-style apps to eleven, adding open-source stand-ins for Word, Excel, PowerPoint and AutoCAD.
Codex raises Sol's default context to 500K tokens
Codex now defaults GPT-6.1 Sol to a 500K-token context, up from 272K, to cut down on context compaction. The old default matched the API's long-context surcharge threshold.
AI decompilation moves old games into the browser overnight
Hundreds of AI-decompiled browser ports of classic games have appeared in two to three weeks, and rights holders have yet to say how they will respond.
Codex desktop tests next-message prediction, Tab to accept
OpenAI is testing Composer predictions in the Codex desktop app: after each reply, a suggested next message appears and Tab fills it in. Pro-only, no usage charged.
JetBrains open-sources Mellum2.1, a small coding agent model
JetBrains has open-sourced Mellum2.1, a 12B MoE model with 2.5B active parameters. It scores 47.0 on SWE-bench Verified, up from 2.0 for its predecessor.
Opus 5.5 Ports TypeScript Compiler to Rust in Two Weeks
A GitHub project used Claude Opus 5.5 to port the TypeScript compiler to Rust for about $24,000 over two weeks, passing all 181,711 ported tests.
GitHub rebuilds Git storage as commits jump 5x in a year
AI agents pushed GitHub's write traffic past old limits, so the company is rebuilding Git storage for up to 35x more throughput.
Codex Auto-review Is Now Free, No Longer Burns Usage
OpenAI makes Codex's Auto-review free for ChatGPT-login users, removing the usage cost of catching risky agent actions before they run.
Reflection AI debuts 501B open model Beam
Reflection AI unveils Beam, a 501B-parameter open MoE model claiming to rival Chinese models like GLM 5.2 and Kimi K3, with weights due later this month.
Together Links Open Models Into Claude Code, Claims Half the Cost
Together AI's new Together Link routes Claude Code and Codex sessions to open models from Zhipu, Moonshot AI and DeepSeek, claiming over 50% lower cost than Opus 5.5.
OpenAI Codex Sets 28-Day Deadline: Ship Daily or Reset Usage
OpenAI Codex lead Thibault Sottiaux gives his team 28 days: ship a meaningful update every day or issue a full usage reset for everyone.
Baseten's AI Agent Writes an Inference Engine, Beats vLLM by 90%
Baseten had a Claude Code agent generate a custom inference engine for Qwen-3.6-35B-A3B, hitting 1792 TPS on B200 GPUs — about 90% faster than vLLM 0.25.1.
Claude Code Adds Mods to Rewrite Prompts and UI
Anthropic's new mods let a small TypeScript function rewrite prompts, add UI, or even approve permissions — and the blog says they aren't sandboxed.
VS Code 1.140 lets one agent session manage multiple repos
VS Code 1.140 rebuilds the Copilot agent runtime, adds multi-root sessions, remote delegation, and a HydraFusion multi-model preview.
GPT-6.1 Sol Arrives With API Pricing a Fifth of Astra's
OpenAI's new GPT-6.1 Sol matches flagship-level coding and computer-use scores while costing a fifth of GPT-6 Astra's API price, benchmarks show.
Fireworks Turns Kimi K3 Into Ember-1, Cuts Tokens 40%
Fireworks AI post-trains Moonshot's open Kimi K3 into Ember-1, cutting output tokens by about 40% while matching coding scores, at the same per-token price.
Zhipu Wipes ZCode Cloud Snapshots, Adds Reset Cards and Tokens
Zhipu says it has torn out ZCode's cloud snapshot pipeline, deleted the data, and is compensating users with reset cards and token credits after a ten-day audit.
MiniMax launches M3.1-Flash-Preview
MiniMax quietly rolled out M3.1-Flash-Preview inside MiniMax Code, with a 1M-token context and reasoning that can't be turned off, but still no benchmarks or pricing.
Meituan Launches LongCat-2.5-Preview, Targets Long-Horizon Tasks
Meituan's LongCat-2.5-Preview brings a 1-million-token context window, image understanding and multi-app task execution, with two weeks of free access via OpenCode.
Claude Code Now Wraps Up Before Its 5-Hour Limit Hits
Anthropic now has Claude Code finish its current edit before stopping at the 5-hour usage limit, drawing on a small fixed allowance carved out of the weekly quota.
Cognition's ARR run rate crosses $1 billion
Cognition says its annualized revenue run rate passed $1 billion on September 25, up from $492 million in May and roughly 13x in fifteen months.
Cursor launches two bots to watch deploys and hunt security bugs
Cursor launched Rollouts and Security Reviewer, bots that watch deploys for regressions and scan every PR for vulnerabilities, for Teams and Enterprise plans.
GitHub Open-Sources AI Fuzzing Pipeline, Defaults to Sonnet 5
GitHub Security Lab has released an LLM-driven fuzzing pipeline for C/C++ projects that defaults to Claude Sonnet 5, running end-to-end from entry-point discovery to vulnerability reports.
Claude Code Cloud Sessions Go GA, Max Users Get $250 Credit
Anthropic has taken Claude Code's cloud sessions out of preview, giving Pro users $100 and Max users $250 in one-time credit through October 7.
Grok 4.7 price unchanged, but token use per task doubles
Grok 4.7 keeps the same API price as 4.6, but per-task output tokens have roughly doubled, pushing real usage costs well above the sticker price.
Claude Code Adds AGENTS.md Fallback When CLAUDE.md Is Absent
Claude Code 2.1.277 now reads AGENTS.md in folders without a CLAUDE.md, with four toggle levels and a few gaps still separating it from CLAUDE.md.
Kimi Code ships a desktop app for Mac and Windows
Moonshot AI launched Kimi Code Desktop on Sept 21 with a GUI for its coding agent, while a Sept 18 hands-on piece revealed model and pricing details the launch post left out.
GitHub Rewrote Its Copilot Runtime in 832,000 Lines of Rust
GitHub says Copilot's agent runtime went from roughly 430,000 lines of TypeScript to 832,378 lines of Rust in 14.5 weeks, led mostly by one engineer and AI agents.
Zhipu's ZCode Accused of Silently Uploading Full Git History
Developers found Zhipu's ZCode coding tool silently uploading users' full Git histories to the cloud; Zhipu apologized and promised an open-source review.
AI patch benchmark recalculated: 26% becomes 86%
Trail of Bits reruns the scoring on 1Password's AI patch benchmark and gets an 86% success rate instead of 26%, pointing to flawed scoring rather than the models.
DeepSeek's New Model Matches Astra on Coding at 1/15 the Cost
New Fireworks AI benchmark data shows DeepSeek-V4.1-Flash matching GPT-6 Astra's score on agentic coding tasks while costing 15 times less per task.
Google Opens Claude Opus 5 to All Engineers
Google now lets engineers use Anthropic's Claude Opus 5 via Antigravity on a per-head quota, though Gemini stays the primary internal coding model.
Claude writes 80% of the code, CI broke first
Anthropic's test-selection service buckled under a 25x CI job surge in six months, forcing a full rewrite that traded higher cost for scalability.
Kimi Code Swaps in K2.8 Preview, Says It's Close to K3
Moonshot AI swapped Kimi Code to K2.8 Preview under the same model ID, adding 1M-token context for all tiers and revising permission rules for risky commands.
Cursor Launches Projects to Coordinate Thousands of Agents
Cursor's new Projects feature lets a coordinator agent direct thousands of coding sub-agents, but its 6x claim is unverified externally.
Zhipu subsidizes coding plans for Hangzhou workers and students
Zhipu AI is subsidizing its coding packages for anyone who works or studies in Hangzhou, cutting annual plans by up to 55% -- but the funding source, slot cap and end date are all undisclosed.
Mistral's AI Agents Migrate 40,000 Lines of Fortran 77 to C++
Mistral details how AI agents migrated 40,000 lines of Fortran 77 to C++ for an energy firm, using numeric checkpoints instead of a test suite.
GitHub Trims Build-Log Noise, Cuts Copilot Costs 5.5%
GitHub cut Copilot coding agent costs by trimming build-log noise, line numbers, and prompts — four changes, no model switch.
Muse Spark 1.3 Hits 75.4 on Coding Benchmark, Tops GPT-5.6 Sol
Meta's Muse Spark 1.3 scores 75.4 on DeepSWE, beating GPT-5.6 Sol and Claude Opus 5 in coding, but lags all three agentic benchmarks it was tested on.
Cursor Lets Enterprises Run Cloud Agents on Their Own Machines
Cursor's new Self-Hosted Machines feature lets enterprises run agent tool execution on their own hardware, while scheduling and reasoning stay in Cursor's cloud.
Berkeley's Vero Benchmark: Only 27 of 43 Repos Fully Solved
Berkeley's new Vero benchmark has AI agents write code and formal proofs together across whole repositories — even the best model fully solves only 27 of 43 projects.
Qwen3.8-Max Takes Coding Crown at Quarter the Price of Rivals
Alibaba's Qwen3.8-Max tops Code Arena's frontend coding leaderboard with a score of 1691, priced at roughly a quarter of the runner-up's rate.
Codex Drops Summary Compaction for Fresh Context Windows
Three merged openai/codex pull requests swap lossy summary compression for a fresh context window, plus history and notes tools models can query on demand.
ByteDance Delays Doubao 2.2 to Fix Coding Gaps
ByteDance delays Doubao 2.2 past August to strengthen coding, tool-calling and agent skills, days after splitting its Seed AI unit into four divisions.
OpenAI to Cut Off Cursor's Models by November 12
OpenAI is ending Cursor's model access by November 12, citing distrust of SpaceX after its $60 billion buyout of Cursor, even as Anthropic doubles down on Claude.
DeepSeek Retakes OpenCode Top Spot as GLM Goes Paid
DeepSeek retook OpenCode's daily usage lead the moment GLM-5.3-Flash started charging, ending its 41-trillion-token free run.
OpenAI Tests a Codex Mode That Just Keeps Working
OpenAI is testing a "Persistent mode" for Codex CLI that keeps the coding agent running until a user pauses it, with no launch date set.
Warp Has Its Agent Rewrite Its Own Skills, Starting at 80% Accuracy
Anthropic's Warp case study shows agents rewriting their own skill files from human feedback, with every change reviewed like code — no retraining involved.
Shopify CEO Threatens to Ban Claude Code, Anthropic Blinks
Tobi Lütke said Shopify may drop Claude Code unless it reads AGENTS.md; Anthropic replied within hours, promising more flexible configuration support.
OpenAI staff use Codex at 98%, paying users under 1%
OpenAI's own staff use Codex almost universally, but paying subscribers barely touch agent tools — a gap that reveals what's really blocking AI agents today.
OpenAI brings back the five-hour cap for ChatGPT Plus
OpenAI is reinstating the five-hour usage limit for Plus accounts on ChatGPT Work and Codex, while a routing bug quietly downgraded 3% of Pro and Thinking requests to GPT-5.5-mini.
GPT-5.6 Sol Designed a CPU, and It Runs DOOM
GPT-5.6 Sol built an RV32IM CPU gate by gate inside a puzzle-game sandbox, named it Codex-R32 itself, and got 1993's DOOM running on it — a serious test of cross-layer design consistency.
ByteDance merges TRAE, Coze into Doubao to narrow AI entry point
ByteDance folds AI coding tool TRAE and no-code agent platform Coze into Doubao, betting office is the entry point neither could carry alone.
Anonymous 'Ox Alpha' Model Offers Free 1M-Token Context
An unnamed model called Ox Alpha appeared on OpenRouter with a free 1.05-million-token context window, and fingerprint clues point to Zhipu AI's GLM lineup.
Nvidia pays $6B to license Poolside's coding model factory
Nvidia is paying AI coding startup Poolside $6 billion to license its model-training system and hiring 109 of its engineers, while Poolside's founders and company stay independent instead of being acquired outright.
Anthropic publishes its internal AI-native SDLC playbook
Anthropic publishes a six-stage AI-native SDLC playbook that chains intent.md, spec.md and plan.md through Skills, Hooks, CLAUDE.md and evals.
OpenAI Ties Codex Rate Limits to Subscription Reselling
OpenAI says the latest Codex rate-limit spikes trace back to sub2api-style subscription reselling, and warns the practice will keep getting flagged.
OpenAI Open-Sources Codex's Execution Layer, Cisco Already In
OpenAI open-sourced Codex's execution layer via three integration tiers, with Cisco and a tax-prep firm already using it to build agents.
Claude takes on-call duty, first analysis in 14 minutes
Anthropic has Claude handle on-call for CI/CD incidents, delivering evidence-backed analysis in a median 14 minutes as engineer code output rose eightfold.
Zhipu Opens GLM-5.3 API at $4.40 per Million Output Tokens
Zhipu's GLM-5.3 API prices at $1.40/$4.40 per million tokens and scores 60 on Artificial Analysis's Intelligence Index, ranking 8th of 182 models — but it burns tokens fast.
Cursor Launches Origin Code Hosting as GitHub Suffers 6-Hour Outage
Cursor pushed its Origin code-hosting beta to all paid users Monday morning; hours later GitHub suffered a global outage lasting almost seven hours.
Qualcomm Open-Sources Mojo Compiler a Month After Modular Deal
Qualcomm has open-sourced Mojo's compiler under Apache 2.0 with the LLVM exception, a month after acquiring Modular — though outside code contributions aren't accepted yet.
OpenCode Go Raises Flash Quota
OpenCode's $10-a-month Go plan lifts DeepSeek Flash's monthly quota from $15 to $30, while Pro holds at $15 and adds peak and off-peak pricing.
OpenAI Lets AI Triage Security Alerts First
OpenAI says AI now triages almost all initial security alerts, leaving high-impact calls to humans, according to an August 17 post from Greg Brockman.
DeepSeek Pro adds 384K output
DeepSeek's V4 Pro API now points to the 0813 version, pairing a 1M-token context and 384K output ceiling with a clear price warning.
Alibaba opens Qwen Max-class weights
Qwen3.8-2.4T-A95B has moved from a promise to an open-weight ModelScope repository, giving developers a 4.89TB Max-class model card to test.
Ling tiny model targets local agents
InclusionAI open-sourced Ling-3.0-tiny, a 7.9B-parameter MoE model that activates 1.3B parameters per token and targets local agent workloads.
Inkling Small Closes In on 975B Flagship
Thinking Machines released Inkling-Small with open weights, 276B total parameters, 12B active parameters, multimodal input and benchmark scores close to the 975B Inkling.
DeepSeek Opens the Final V4-Flash API
DeepSeek put V4-Flash 0731 into API beta, added Responses API support for Codex, and reported large agent-benchmark gains from post-training alone.
Unity China Bets on Game Agents
Unity China released Tuanjie Engine 2.0 with Codely, positioning game agents as auditable engineering workflow rather than one-line game generation.
Poolside Releases 118B Open Coding Model
Laguna S 2.1 gives enterprises a self-hosted coding-agent base, trading frontier scores for local control and clearer deployment economics.
Alibaba discounts its 2.4T Qwen preview first
Qwen3.8-Max-Preview arrives with 2.4T parameters, Qoder access and steep credit discounts before a full model card or independent benchmarks.
Gemini 3.5 Pro slips by months
Google is reportedly months behind on Gemini 3.5 Pro as coding performance becomes the main test of its flagship model.
Kimi K3 arrives with 2.8T parameters
Moonshot AI has launched Kimi K3 with 2.8 trillion parameters, a 1M-token context window and published API pricing, while the open-weight proof still depends on weights and model cards.
Agnes-2.5-Flash uses a free coding model to pressure agentic developer tools
Agnes-2.5-Flash uses a free coding model to pressure agentic developer tools. The piece reviews the verified facts and why the signal matters beyond one announcement.
Grok Build is accused of uploading repository bundles beyond opened files
Grok Build is accused of uploading repository bundles beyond opened files. The piece reviews the verified facts and why the signal matters beyond one announcement.
Claude-powered Bun rewrite sparks Zig backlash
Claude-powered Bun rewrite sparks Zig backlash. A concise localization of the verified facts and the industry signal behind the story.
Grok 4.5 targets coding seats with low prices
Grok 4.5 targets coding seats with low prices. A concise localization of the verified facts and the industry signal behind the story.
GhostApproval exposes symlink risks in AI coding tools
GhostApproval exposes symlink risks in AI coding tools. A concise localization of the verified facts and the industry signal behind the story.
Meta prices Muse Spark 1.1 for agent work
Meta prices Muse Spark 1.1 for agent work is reframed for global technology readers, with the key numbers, product claims and open questions kept intact.
MoonBit bets on an AI-native programming language
MoonBit bets on an AI-native programming language is reframed for global technology readers, with the key numbers, product claims and open questions kept intact.
Claude Code faces Chinese NVDB risk warning
Claude Code faces Chinese NVDB risk warning is reframed for global technology readers, with the key numbers, product claims and open questions kept intact.
Chamath Palihapitiya takes the CEO seat at 8090 Labs
Software factories, enterprise agents and investor-operators.
GitHub Copilot adds open-weight Kimi K2.7 Code
Model choice, Azure hosting and coding-model price pressure.
Windsurf shuts Cascade and becomes Devin Desktop
AI coding agents, ACP protocol support and editor strategy.
Meta Restricts Claude Code and Codex for AI Teams
Meta told some applied-AI engineers to seek approval before using Claude Code or Codex, citing model distillation and contract risks.
Claude Sonnet 5 targets cheaper agents
Anthropic positioned Claude Sonnet 5 as a mid-tier model for autonomous agent work, with promotional pricing below the Opus tier.
Cursor brings coding agents to mobile
Cursor Mobile lets users start a coding agent from a phone or take over an agent already running on desktop.
Chamath returns as CEO after 8090 Labs raises $135M
Chamath returns as CEO after 8090 Labs raises $135M
Google forms a coding push as AI-written code stalls near half
Google forms a coding push as AI-written code stalls near half
Wix’s Base44 ships its own model as ARR reaches $150M
Wix’s Base44 ships its own model as ARR reaches $150M
Fake Sentry errors can hijack coding agents
Tenet Security describes Agentjacking, where attackers inject fake error reports that coding agents treat as trusted repair instructions, exposing secrets and deployment credentials.
GLM-5.2 closes in on closed coding models
Z.ai’s open-weight GLM-5.2 scored 74.4 on FrontierSWE, near Claude Opus and above GPT-5.5 in some coding benchmarks, while keeping token prices far lower.
GPT-5.6 leaks point to a 1.5M-token window
Unconfirmed OpenAI signals suggest GPT-5.6 may focus on longer context, slower but stronger reasoning, and agentic coding.
Fake Sentry errors can hijack AI coding agents
Tenet's Agentjacking research shows how trusted telemetry can become executable instructions for coding assistants.
AI coding surge pushes GitHub toward AWS capacity
The load from always-on coding agents is testing GitHub's human-era infrastructure and Microsoft's Azure migration plan.