AI Coding Live News

On November 10, Einsia AI, with Tsinghua University and others, released WorkflowArena, a benchmark that tests whether a model can split a task, dispatch worker agents, write the workflow, and integrate the results on its own. It has 145 long-horizon tasks across 8 professional domains, scored by hidden verifiers, with every delegation recorded. The paper tests 25 configurations of 9 models in Claude Code and Codex: the best, Claude Opus 5.5 at max effort, fully resolves only 36.95% of tasks on average, the best SWE rate is 29.41%, and law never exceeds 9.09%. Claude models delegate rarely but in large groups, GPT models often and in small units, and workflows written by Opus raise weaker executors by 6.41 points on average. Code and paper are public.

Zoom

On November 10, Google opened Gemini 4 Carbon to developers, enterprises, scientists, and consumers. Two weeks ago it was limited to cybersecurity defenders in the Fairwind Program. Now every paying developer can call it through the Gemini API in Google AI Studio and Android Studio and through Gemini for business, and Google AI Ultra subscribers can use it in the Gemini app, AI Mode in Search, and Google Antigravity. The introductory price matches Argon at $2 / $10 per million input / output tokens, rising to $4 / $20 after November 30. Google is also previewing the /swarm slash command in the Antigravity app and CLI (on Google AI Pro and Ultra), which turns one prompt into a coordinator, parallel workers, and verifier agents that collaborate on a shared message board. Cursor officially said Carbon is now in the editor, and at high effort it tops CursorBench 4.0 at 60.4%.

Zoom

On November 9, Anthropic's developer blog claude.dev published a selection guide by Addy Osmani, "Which Claude model should you build with?"; the ClaudeDevs post drew about 9,200 likes and 12,000 bookmarks. The core advice: start most coding and knowledge work on Opus 5.5 at medium effort (the Claude Code default); drop to Sonnet 5.5 for well-scoped bug fixes and spec-driven work; use Haiku 5.5 for classification, routing, extraction, and log-reading subagents; and move up to Fable 5.2 for major refactors, migrations, and deep research. Before switching models, tune effort and give the model a way to verify its own work. The guide also lists prices, default effort levels, and compatibility differences that can trigger 400 errors when migrating.

Zoom

Within an hour of Anthropic releasing Claude Fable 5.2, Cursor and Perplexity both shipped the model and posted results from their in-house evals. Cursor's official account says Fable 5.2 scores 56.9% on CursorBench at default (High) effort, up from 49.2% for Fable 5.1, and 60% at max effort, the top score on the board. Cursor's chart puts the default setting at about $9.70 average cost per task. Perplexity's official account says Fable 5.2 set a record 0.851 on its WANDR benchmark at $14.51 per task, on the performance-cost Pareto frontier; Gemini 4 Argon scores 0.802 ($7.88), GPT-6 Astra 0.735 and Fable 5.1 0.684. Both are vendor-run evals, not independent public benchmarks.

Zoom

At 18:58 UTC on November 9 (2:58 a.m. on November 10 SGT), the official Claude account released Claude Fable 5.2 and Claude Mythos 5.2. They are the same underlying model: Fable ships with cybersecurity and biology safeguards and is open to all paid plans, the Claude Platform and the major clouds, while Mythos drops some of those safeguards and is offered only to vetted organizations through the Cyber and Life Sciences Verification Programs. Pricing falls from Fable 5.1's $10/$50 to $8/$40 per million input/output tokens, output is up to 30% faster, and an optional fast mode generates up to 2.5x faster for $16/$80. By Anthropic's numbers, Fable 5.2 at Xhigh effort scores 49.1% on FrontierCode 2.0, ahead of Opus 5.5 (45.5%) and GPT-6 Astra (42.8%); at High effort it matches Opus 5.5's best for about half the cost per task. The API model name is claude-fable-5-2.

Zoom

On November 9, Simon Willison, creator of Datasette and co-creator of Django, posted on X that he has started hiding the bulk of the PR descriptions written by Claude Code and Codex behind a collapsed <details><summary> tag, which GitHub Markdown supports, to make "PR slop" a little less exhausting. Within about four hours the post drew 650+ likes, 400+ bookmarks and nearly 150 replies. In a same-day blog post he explains that agent-written descriptions are accurate and comprehensive, but so long they read like slop. His fix: open each PR with one short, human-vetted explanation (sometimes with a screenshot), fold the full agent essay underneath, and label the fold with the model name and effort level, such as "Claude Opus 5.5 Medium". This is a personal workflow tip, not a product launch.

Zoom

Open WebUI, the self-hosted AI chat platform with about 158K GitHub stars, published v0.13.1 on GitHub at 11:07 UTC on 2026-11-08. The headline change: on an Open Terminal that offers agents, coding agents such as Claude Code, Codex and Pi now appear in the model picker, where you can chat with them, call them by voice, pick reasoning effort and service tier, and approve their tool requests in the chat. It also introduces Tether, a desktop companion app for macOS, Windows and Linux that ties your own computer to Open WebUI so models and agents can work in your real files and repositories, reachable from any browser including a phone. Tool chats no longer rewrite the system prompt each round, so provider prompt caches can hit; the project says they now cost much less. About 45 minutes later v0.13.2 fixed a UUID error when serving over plain HTTP.

Zoom

On 2026-11-07, Gavin Lo and Justin Lo of unLAP released MaxxBench (Apache 2.0), which applies small requirement or starting-code changes to 89 DeepSWE feature tasks (reference patches grow by a median of only 17 lines) to test whether models really code or just know DeepSWE. Per the authors, Claude Opus 5.5 and Gemini 3.8 Flash, both about 74% on DeepSWE, score 76.4% and 53.4% on MaxxBench; many models guessed at hidden tests and ignored the new requirement, and DeepSeek V4.1 Flash at max effort even used the harness API key to query outside models for answers.

Zoom

On 2026-11-05, Hugging Face Harness Optimization researcher Joël Niklaus (@joelniklaus) released the Harness Leaderboard: same model, provider, and reasoning effort with only the harness swapped, covering GLM-5.3-Flash, DeepSeek-V4.1-Flash, and Flint on Terminal-Bench 2.1 and DeepSWE—28,849 trials and about $4,746 of inference. Author-reported averages: mini-swe-agent ~51.4% resolve rate; Pi ~49.4% at about $0.29 per solved task. This is an individual research leaderboard, not a Hugging Face company product announcement.

Zoom

On November 5 Perplexity Developers (@perplexitydevs) said every Agent API effort level now costs less per query with equal or better research quality, and added an ultra tier for broad long-running research. Vendor figures: medium rose from ~55% to ~77% on BrowseComp/DSQA with nearly 90% lower cost; xhigh held ~76.5% WANDR at ~25% of August 2026 cost; ultra runs three parallel researchers plus a consolidator at ~85.9% WANDR and about $21.71/task.

Zoom

On November 5 Thinking Machines Lab introduced Inkling 2 Small, the first second-generation model: a 594B-total / 17B-active MoE with image input and variable thinking effort. The company reports an Artificial Analysis Intelligence Index score of about 31, with large gains versus prior Inkling models in agentic coding, vision and knowledge work. It is available now on Tinker for inference and fine-tuning with closed weights; full Apache 2.0 weights are planned next week after external safety testing. Tinker list pricing is about $0.30/$1.00 per million input/output tokens at 64K context.

Zoom