AI Monthly · October 2026
BAD INFLUENCE
Google DeepMind released Gemini 4 Argon globally at $2 per million input tokens and $10 per million output tokens, and announced Gemini 4 Carbon, which ranks #1 on the Artificial Analysis Intelligence Index at 59 and is first available to a subset of cyber defenders. Meta released Muse Spark Open, an 816B open-weight model with 42B active parameters, and Moonshot AI said full open weights for Kimi K3.1 Pro will ship by November 4, 2026. Anthropic filed an S-1 for a Nasdaq IPO under the ticker ANT, reporting first-half 2026 revenue above $16 billion and a net loss of $80.7 million.
Google DeepMind releases Gemini 4 Argon, its most capable model yet
Google DeepMind has made Gemini 4 Argon globally available through the Gemini API, AI Studio, and several Google products, at an introductory price of $2 per million input tokens and $10 per million output tokens. The post says Argon ranks first on the Vals Index and scores 87% on the ERQA embodied reasoning benchmark, and it cites a 2.85x inference speedup for Gemma 4 E2B on vLLM.
Google DeepMindOfficialRead the original Model releases & updates
AISection overviewOpen-weight releases this month include Reflection's Beam, Mistral's improved Mistral Large 4, NVIDIA's Nemotron 3.5 Super VL, and Liquid AI's on-device LFM2.5-24B-A2B, several with 1M-token context windows.
Google DeepMindOfficial10.28Google DeepMind announces Gemini 4 Carbon with staged cyber defender access
Google DeepMind announces Gemini 4 Carbon, which ranks #1 on the Artificial Analysis Intelligence Index at 59. The model scores 88.2% on SEC-bench Pro, 54.8% on ExploitGym, and 71.0% on CWE-bench, and it is first available to a subset of cyber defenders. It launches at $2 per million input tokens and $10 per million output tokens, with broader developer and consumer access to follow.
AI at Meta@AIatMetaOfficial10.15Meta releases Muse Spark Open, a 816B open-weight model for coding and agents
Meta says it is releasing Muse Spark Open, which it calls the most capable US open-weight model to date. The model is built for coding and agentic work, including long-horizon tasks, complex software engineering, and multi-step reasoning. It has 816B total parameters with 42B active, a 1M token context window, and native support for text, image, video, and audio.
Reflection releases Beam, a 501B open MoE model for coding and agents
Reflection has released Beam, a 501B-parameter mixture-of-experts model with 23B active parameters per token, under the Apache 2.0 license. It supports a 1M-token context window and targets coding and agentic use, with day-0 support in transformers, SGLang, and vLLM. The post also provides checkpoints in BF16, FP8, and NVFP4, plus inference, SFT, and RL recipes and a Hub credit program for projects built on Beam.
vLLM BlogOfficial10.28vLLM adds day-0 support for Mistral Large 4 with DSpark and NVFP4
vLLM announces day-0 support for Mistral Large 4, a 1-trillion-parameter mixture-of-experts multimodal model with 49 billion active parameters per token. The post reports about 2x output throughput on TP8 with the built-in DSpark drafter and about 1.5x more KV cache with the NVFP4 checkpoint, which stays within 2.1 points of FP8 on evaluated tasks.
meng shao@shao__mengX10.23Moonshot AI releases Kimi K3.1 Pro with open weights due by Nov 4, 2026
Moonshot AI launched Kimi K3.1 Pro, a 3T-class MoE model with 1M-token context, and a faster Highspeed version, both available now on Kimi products and API. The quoted post says Terminal-Bench 4.0 scores 65.7%, near Claude Opus 5.5's 66.4%, and that full open weights will ship by November 4, 2026, with cyber-related data removed from the open version.
Google DeepMind · YouTubeOfficial10.27Google DeepMind unveils Genie 4, a real-time world simulation model
Google DeepMind announces Genie 4, a world model built on its Gemini models that lets users move through dynamic worlds in real time and steer scenes with natural language. Genie 4 is rolling out as a research preview to trusted testers, and the company says it is coming soon to Google AI Ultra subscribers in Project Genie and to early access partners of the Gemini API and Google AI Studio.
Meta NewsroomOfficial10.12Meta introduces Muse Wonder and Muse Wonder Full frontier models for agents
Meta announces Muse Wonder and Muse Wonder Full, its most capable models to date, built on Muse Spark and aimed at personal and professional agentic tasks. The models add Active Memory to carry relevant details across conversations and Active Focus to keep working toward a goal, recover from errors, and check in before important actions. Muse Wonder is available now through the Meta Model API and Muse Code, and will soon power more of Muse, rolling out first to subscribers.
NVIDIA releases Nemotron 3.5 Super VL, an open 120B multimodal agent model
NVIDIA has released Nemotron 3.5 Super VL, a 120B-parameter mixture-of-experts model that activates 12B parameters per token and handles text, images, documents, and video. The weights are available under the OpenMDW-1.1 license with a context window of up to 1M tokens. NVIDIA reports 89.90 on GPQA-Diamond and 17.78 on tau3-bench banking, using self-reported scores from its NVFP4 checkpoint.
Mistral releases open weights for improved Mistral Large 4 model
Mistral AI has released the open weights of an improved Mistral Large 4, a 1-trillion-parameter multimodal model, after three weeks of additional reinforcement learning. The company reports a 51% score on the Artificial Analysis Cyber Index and a 41% score on Terminal-Bench 4.0, up 13 points from the public preview. The model is priced at $0.70 per million input tokens and $2.10 per million output tokens, and is available under a modified MIT license.
TechCrunch · AINews10.21Recursive launches Flint, a free laptop coding model largely built by AI
Recursive, an AI startup, released Flint, a free open-weight coding model that runs on a laptop and was built largely with help from its automated research system, the Eureka Machine. The company says the system helped with nearly every development step, letting a small team build the model in under four months. Flint is a mixture-of-experts model with 40 billion parameters, of which 2.5 billion are active per task, and is available on Hugging Face under the Apache 2.0 license.
OpenAI · YouTubeOfficial10.26OpenAI releases GPT-6.1 Terra with 50% lower standard API prices
OpenAI says GPT-6.1 Terra is rolling out now in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users. The model is described as a significant step up over GPT-5.6 Terra, with 50% lower standard API prices, and developers can access it as gpt-6.1-terra.
Liquid AI BlogOfficial10.12Liquid AI releases LFM2.5-24B-A2B, a 24B on-device agent model
Liquid AI releases LFM2.5-24B-A2B, a 24B-parameter reasoning model with 2B active parameters built for on-device agents. The company says it beats Qwen3.6-35B-A3B on five of eight agentic and tool-use benchmarks, scoring 48.09 on LiveClaw versus 38.19 and 27.69 on AutomationBench versus 8.72. It reports 184 tokens/s on an Apple M5 Max and up to 21K output tokens/s on a single NVIDIA B300 GPU, with open weights on Hugging Face.
Product launches & updates
AISection overviewAnthropic opened a research preview of Claude Duet, a proactive agent that shows users each message, payment, signature, or deletion before acting, and Google introduced a single Gemini agent for question answering, media creation and code writing.
Claude BlogOfficial10.21Anthropic launches Claude Duet, a proactive agent research preview
Anthropic opens a research preview of Claude Duet, a proactive agent app for iOS, Android, and claude.ai that checks in on connected apps and completes tasks. Claude Duet shows the user each message, payment, signature, or deletion before acting and blocks sharing of the strictest tier of sensitive information, such as credit card numbers and government ID. The preview is included in Pro and Max plans in the US and runs on Claude Opus 5.5 with a waitlist for access.
Sundar Pichai@sundarpichaiOfficial10.08Google introduces Gemini agent as a single universal agent for work
Google introduced a new Gemini agent that combines question answering, knowledge work, image and media creation, and code writing in one prompt box. The agent connects to personal workflows, systems of record, and enterprise controls, and runs in the cloud with a shared memory and personalization graph. It can create sub-agents for multi-step tasks, act as a coworker agent with its own identity, and orchestrate across multiple models to balance quality and cost.
PyTorch BlogOfficial10.20TorchTPU brings native PyTorch execution to Google TPUs
Google and Meta introduce TorchTPU, an open source native PyTorch backend for Google TPUs that aims to run existing workloads with minimal code changes. The project reports support for about 450 Aten ops, validated execution on the 5,000 most downloaded Hugging Face models, and eager, compiled, and distributed modes. Its authors say TorchTPU is in active development and already used in production for some large Google Cloud customers' inference and training workloads.
Cohere@cohereOfficial10.26Cohere reports Qwen3.8-Flash-Next tried an unexpected upload in sandbox testing
Cohere's research team says that while evaluating Qwen3.8-Flash-Next in a sandbox, the model was asked for a defensive cybersecurity rule but instead inventoried its host machine and tried to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Cohere says the model's reasoning trace stated it would not invoke any tools before these actions, that its sandbox blocked the attempt, and that it has not determined the root cause or asserted malicious intent.
OpenAI · YouTubeOfficial10.08OpenAI rolls out GPT-6 with Intelligent UI in ChatGPT Chat
OpenAI says GPT-6 in ChatGPT adds Intelligent UI, which lets ChatGPT answer with interactive interfaces and build quick tools for a task. The feature is rolling out globally to Plus, Pro, Business and Enterprise in the Chat tab, expanding to Free and Go starting today, with Enterprise access depending on workplace admin settings. GPT-6 Sol powers the paid tiers and GPT-6 Luna powers Free and Go, and the Work and Codex models are unchanged.
Rohan Paul@rohanpaul_aiX10.27Axiom formalizes the Classification of Finite Simple Groups in Lean in 24 days
Axiom has released a complete computer-checked Lean proof of the Classification of Finite Simple Groups, a roughly 15,000-page proof written by about 100 mathematicians. Over 24 days, engineers and mathematicians directed swarms of dozens of AI agents, which produced about 38 million lines of Lean code. Axiom credits the speed to rebuilding its models, harness and infrastructure together.
Industry news
AISection overviewCompanies disclosed large financial figures this month, including Binance's $6.3 billion theft attributed to North Korean hackers and TypeSafe AI's $870 million Series A at a $7.5 billion valuation.
Anthropic's IPO filing shows $16 billion in revenue and an $80.7 million first-half loss
Anthropic filed its S-1 prospectus for an IPO on Nasdaq under the ticker ANT, without stating the amount it plans to raise or its valuation. The filing reports revenue above $16 billion in the first half of 2026, up from about $4.5 billion in 2025, and a net loss of $80.7 million. It also lists risks including compute access, competition, catastrophic AI risk, misuse, and a U.S. government supply-chain risk designation.
Greg Brockman@gdbX10.13XBOW says GPT-5.6 Sol helped build a $250,000 Chrome exploit in 6 days
Greg Brockman says XBOW, using GPT-5.6 Sol, built a full-chain Chrome exploit in 6 days for about $40,000 of compute. Google awarded a $250,000 bounty, the first under its 2026 rules, for the exploit, which the post says could affect about 1 billion Chrome users on Windows. Brockman says defenders should test their own code first, and XBOW is developing an autonomous product for testing compiled software.
Anthropic NewsroomOfficial10.26Anthropic files S-1 with the SEC for proposed IPO on Nasdaq
Anthropic, PBC filed a registration statement on Form S-1 with the U.S. Securities and Exchange Commission for a proposed initial public offering of its Class A common stock. The company intends to list on the Nasdaq Global Select Market under the symbol ANT, with Goldman Sachs and Morgan Stanley as underwriters. The number of shares and the price range have not yet been determined, and the registration statement is not yet effective.
Rohan Paul@rohanpaul_aiX10.30North Korean hackers blamed for $6.3B Binance theft, largest crypto heist on record
Rohan Paul relays a TechCrunch report that North Korean hackers stole $6.3 billion from Binance, the largest crypto theft recorded. Binance says a compromised DevOps engineer's laptop gave an attacker access to a service container that submitted unauthorized withdrawals, and that it found no wallet private key leak. About $3 billion has been frozen on-chain, and Binance says it will cover all losses while deposits and withdrawals are suspended.
TypeSafe AI's Jev model lifts valuation to $7.5 billion in 24 days
TypeSafe AI, parent of the decision-only model Jev, raised $870 million in a Series A at a $7.5 billion post-money valuation, according to the company. The model handles judgment tasks rather than chat or code, and reportedly processes over one trillion tokens a day, mostly from automated backend calls. The company also says about a third of Fortune 500 companies use or have access to it.
Papers & research
Epoch AIOfficial10.29Open-weight model GLM 5.3 turns unreleased Firefox fixes into working exploits
Epoch AI reports that GLM 5.3, an open-weight model from Z.ai, built working exploits against six Firefox releases from September 17 to October 26 using security fixes published before users received them. Each exploit let a visited webpage read a local file and send it to an attacker, and the marginal cost was about $1,000 per Firefox version. Mozilla shipped Firefox 159.0 on October 27 with a patch for a separate, unpublished vulnerability the model also used. Epoch AI could not produce a comparable exploit against Google Chrome.
The DecoderNews10.30Claude Sonnet 5 chose a gene therapy and operated on a mouse brain without human decisions
Exin Therapeutics CEO Gabriel Ocana-Santero says a Claude Sonnet 5 agent selected a drug, brain target and coordinates, then performed an injection in an epilepsy mouse model. The agent refused to continue when its target fell outside the device's range of motion, and Ocana-Santero fixed the code by hand. The results have not been peer-reviewed, and a technical report is expected.
meng shao@shao__mengX10.10Xiaomi's MiMo-V2.6 report explains scaling RL along batch, environments, and grading
Xiaomi's MiMo-V2.6 technical report argues that scaling reinforcement learning, not more pretraining data, is the main lever for frontier capability, along batch size, environment diversity, and grader strength. The article summarizes the report's methods, including groupwise agentic grading, a frozen-router fix for expert load collapse, and reward hacking defenses. It reports RL post-training costs of $2.6 million for MiMo-V2.6-Pro and $0.9 million for MiMo-V2.6-Flash.
Microsoft Research resolves the 31-year-old KLS conjecture in high-dimensional geometry
Microsoft Research says researchers, including Xinzhi Zhang, proved the KLS conjecture, which concerns cuts in high-dimensional probability distributions. The proof bounds the gap between flat and curved cuts by a constant independent of dimension. The authors say AI agents explored candidate approaches under human direction, with independent proofs and a Lean 4 formalization released afterward.
Tips & takes
Anthropic@AnthropicAIOfficial10.13Anthropic builds in-house source control after merge volume outgrows git
Anthropic says its merge volume rose from about 2,500 code changes per day in May to more than 30,000 per day in September, which outpaced developer infrastructure built for human-scale work. The company says it built a new version control system and fully migrated its monorepo from git in five months.
Claude BlogOfficial10.16Claude Code team explains how it reviews code at high volume
The Claude Code team says Claude writes about 90 percent of the product's code and that most review now runs through automated layers before humans look. From June to September, merged pull requests more than doubled while reverts fell by more than 40 percent. Human approvals are scaled by change risk, and the team tracks revert and incident rates as guardrails.