Model intelligence.
Follow the changes. Understand what comes next.
Google Launches Gemini 4 Neon in Antigravity With 1M Context, Four Thinking Levels, and Benchmarks vs. Claude Opus 5.5 and GPT-6.2 Sol
Announced on November 19, 2026 at 05:51 Beijing time, Google's Gemini 4 Neon features a 1M-token context window, four thinking levels, and planner-executor pairing with Gemini 4 Hydrogen in Google Antigravity, outperforming Claude Opus 5.5 and GPT-6.2 Sol across most of 18 published benchmarks.
Latest stories349
Newest publishedGoogle DeepMind Introduces AlphaProof Code, Formally Verifying Key Signal Protocol Properties at Machine-Code Level
On 10 November 2026, Google DeepMind introduced AlphaProof Code, a formal-verification framework of human experts plus AI agents, not a new chat model. Researchers proved selected key security properties of libsignal down to compiled machine code / AArch64. The paper and artifacts page say about 6.5 million lines of machine-checked Lean proof code; the blog body says over six million lines of Lean. Proofs were produced by Gemini Carbon agents (Gemini 4 Carbon in the paper), not a new Carbon API or pricing release. The blog says just four weeks; the paper, a team of five over four weeks.
OpenRouter Lists Nex-N2.6 Under the ID nex-agi/nex-n2.6
OpenRouter listed Nex-N2.6 at 11:46:19 Beijing time on 10 November 2026 under nex-agi/nex-n2.6. Current rates are $0.12 / $0.40 / $0.024 per million input/output/cache-read tokens; 20% off and struck-through $0.15/$0.50/$0.03 are not live rates. Context is 262,144 tokens. OpenRouter's model card describes a ~321B GLM-5.3-Flash model for computer use, coding, and long-horizon tool tasks; that wording is not from the Nex AGI official website. That site still shows Nex-N2.5, not N2.6. No public Hugging Face card (nex-agi/Nex-N2.6) or official N2.6 chart was found. We haven't tried the model.
Alibaba Cloud Raises Token Plan Credits; Pro Is $43 Internationally and ¥312 in China
Alibaba Cloud began rolling out higher Token Plan Credits at 20:00:00 on 9 November 2026 (UTC+8), timed to the actual update. Both sites offer 20,000/45,000/80,000/200,000 Credits at $6/$10/$18/$43 in Singapore and 39/79/139 yuan (unchanged) plus Pro's official discount of 312 yuan in Beijing. Cycles get the new total; used Credits remain; if exhausted, the difference lasts to the cycle's end. The new Pro price starts next cycle. Credits are not one-to-one with Tokens. The plan is for interactive use, not scripts or batch calls. Night discounts, new-user prices and usage packs are separate.
Anthropic Launches Roadshow for Its Proposed IPO; Expected Price $70.00–$78.00 a Share
On 10 November 2026, Anthropic, PBC launched its proposed Class A IPO roadshow: 825,000,000 shares (company 801,351,354; certain existing stockholders 23,648,646; no company proceeds from those sales). It expects to grant a 30-day underwriter option for up to 123,750,000 additional shares. Expected price $70.00–$78.00 a share; applied to list as ANT on the Nasdaq Global Select Market. SEC registration filed but not yet effective; securities may not be sold, nor offers to buy accepted, until it is. The announcement does not constitute an offer to sell. This follows the 26 October 2026 Form S-1.
Anthropic Updates Its Usage Policy on Cruel Behavior Toward Models, Effective November 12
Anthropic published an updated Usage Policy on 8 October 2026, effective 12 November, not yet in effect as of 10 November 2026. News commentary, not policy text, bars sustained, needless extreme cruelty to models with no discernible purpose, excluding frustration, pushback, dark creative themes, or testing and research. The policy omits "extreme" and "no discernible purpose." Main enforcement, the news post says, is ending rare conversations with persistently abusive users on Claude.ai and Claude Code. Warnings through termination apply; a block is not itself a violation, in a broader refresh.
MiniMax M3.1 Flash Is Free for a Limited Time on OpenCode Console
OpenCode Console lists MiniMax M3.1 Flash Free (minimax-m3.1-flash-free) as free for a limited time. Input, output and cached read are Free; cached write is "-", not Free. The models API lists a 1,000,000-token context and a 512,000-token output limit, with no end date. The provider uses zero retention and does not use the data for training. Free-tier chat models require OpenCode. This is not MiniMax's paid API ($0.15/$0.60 per 1M; cached read $0.015) or OpenCode Go's paid minimax-m3.1-flash. Terms are on the Console Models page; enter via OpenCode Console. We haven't logged in to try it.
Claude Code 2.1.322 Adds Fast Mode on Vertex in the Global, US and EU Regions
Anthropic released Claude Code 2.1.322 on GitHub at 2026-11-10 10:29:43 Asia/Shanghai. The release says it adds fast mode on Vertex in the global, us and eu regions, falling back to standard speed if Vertex refuses it. On the npm package @anthropic-ai/claude-code, latest and next point to 2.1.322; stable remains 2.1.311. We haven't installed or tried this build.
Seedream 5.5 Adds Stable Multi-Turn Editing and Native RGBA Assets
According to ByteDance Seed’s official blog (November 9, 2026), Seed launched Seedream 5.5. Seed says stabler multi-turn editing keeps unedited regions at pixel level for up to 100 rounds on the same image; native RGBA covers full or partial transparency; look and dense small text improve. MagicArena charts show higher usability and MOS vs Seedream 5.0 Pro across six cases, +55.62% realism I2I GSB vs prior, −6.18% vs Nano Banana 2.1 on that cell, and text corruption 0.4% severe / 1.7% minor. Rollout continues on Dreamina, Doubao, Doubao Work, and Volcano Ark; a Pro version will follow.
Antigravity App v2.30.0 Adds Knowledge Work Layout for Research and Writing
According to Google Antigravity’s official changelog, Antigravity App v2.30.0 (November 9, 2026) adds a Knowledge Work layout for research and writing, unavailable on Business and Enterprise accounts. First sign-in can match a work type to a layout, and setups from Claude Code, Codex, or Cursor can move into an owned plugin. Voice mode has finished rolling out except for Business and Enterprise; Remote agents have finished for Google AI Ultra and paid Google AI Pro, after both began in v2.29.0 (November 3, 2026).
Grok Bot Official Account Says Weekly Limits Reset for All Users
Grok Bot's official account @bot said on November 10 that weekly limits have been reset for all users. The post reads: "We've reset weekly limits for all Grok Bot users. Enjoy!" According to verification by our site operator, after this reset users' original natural reset time did not change, unlike OpenAI, which pushes users' natural reset back 7 days after every reset.
OpenAI Introduces Control Room to Track and Halt Research Model Workloads
According to OpenAI (@OpenAI) on X at 2026-11-10 06:01:00 Asia/Shanghai (UTC 2026-11-09T22:01:00.000Z), Safety blog Nov 9, 2026: Control Room tracks research-infra model workloads and can stop them quickly on safety/security concerns. Misalignment monitoring may send a stop (job_id). After a stop, output stops within ~35s (auth-cache lag; goal 0). Already on research infrastructure; external serving has separate takedown. Fail-closed denies inference if Control Room is down—live for all frontier research; other research rolling. Not a ChatGPT kill switch. We haven't tested Control Room.
Thinking Machines Inkling 2 Small Lands on OpenRouter Free Tier
According to the OpenRouter model catalog, Thinking Machines’ Inkling 2 Small is on the free tier as thinkingmachines/inkling-2-small:free (API created 2026-11-09T22:04:07Z=2026-11-10 06:04:07 Asia/Shanghai). As of this check: $0/$0, context 1,048,576 (OR-side only), text+image→text, 17B active / 594B MoE. Thinking Machines’ Nov 5, 2026 news matches the specs (Tinker, closed weights, planned Apache 2.0 after external safety)—launch background, not this listing. OR free ≠ open weights out. Rates/limits can change. We haven't called this model.