Frontier: latest AI news

New models, releases and capabilities from the frontier labs, linked to the model cards and announcements.

2 Nov OpenAI starts merging ChatGPT's Chat and Work tabs into one interface
2 Nov SemiAnalysis launches GDPx, a 206-task expert-work benchmark where the best model scores 12%
2 Nov Du Xiaoman and HKUST launch XuanYuan 4.5, billed as a self-evolving finance model
1 Nov Musk says Grok Bot usage was cut back because serving it cost too much relative to revenue
1 Nov Anthropic's Mythos page briefly named an unreleased Claude Mythos 5.5 before it was taken down
1 Nov Hidden code in Google's Antigravity shows a voice-driven 'Chief of Staff' agent, TestingCatalog reports
1 Nov DeepSeek open-sources a prototype always-on coding agent that takes work from a GitHub repository
1 Nov OpenAI moves ChatGPT and Codex to one shared usage budget as it folds Work into Chat
1 Nov Xiaomi's MiMo-V2.6-Pro, a 1.02T-parameter MoE model, reports 71.9 on DeepSWE and goes live on DeepInfra
1 Nov DeepSeek publishes its Super-agent design and open-sources dsh-superagent, a proactive coding-agent prototype
1 Nov Meituan releases LongCat-2.5, a 1.6T-parameter native multimodal model for long-horizon agent tasks
31 Oct GPT-5.6 Sol, running in Codex, finishes a peaceful-mode Factorio game and launches the rocket
31 Oct Personal AI agent startups route failed tasks such as phone calls to human workers, CNBC reports
31 Oct Apple confirms Apple Intelligence launches in China in December with iOS 27.3, with Alibaba's Qwen
31 Oct OpenAI lists gpt-6-astra-law on its API pricing page at $10 per million input tokens and $50 per million output
30 Oct Anthropic confirms Claude Code prompt rule steering Opus 5.5 away from recreating existing video games
30 Oct humans& releases &Tutor, a tutoring model trained against its Persimmon simulated-learner model
30 Oct Anthropic renames Claude Max 20x to Ultra and drops the 5-hour limit for verified Ultra accounts
30 Oct Google releases Gemini 4 Carbon to selected cyber defenders first
30 Oct OpenAI reorganizes its Model Spec around five core values
30 Oct Mistral releases open weights for Mistral Large 4, its 1-trillion-parameter model
30 Oct OpenAI launches Private Intelligence, bundling zero data retention with a preview of confidential-computing inference
30 Oct Apple CEO John Ternus says Apple does not 'want to ship AI for AI's sake'
30 Oct SenseTime open-sources SenseNova-Loop-9B-Preview, a 9-billion-parameter model that runs 24 of its 32 layers twice
30 Oct Microsoft releases Mage-VL-Cowork, a 4B computer-use agent it reports at 66.0% on OSWorld-Verified
30 Oct DiffuSpace releases Clover-35B-A3B-Preview, an open-weight diffusion language model scoring 73.6 on SWE-bench Verified
30 Oct Baidu says its search now retrains itself from live feedback and cuts its search API to 4.9 yuan per 1,000 calls
30 Oct DeepSeek website code points to an unannounced 'Cyber' security model gated by real-name checks
30 Oct The Biological Computing Co. releases mischief 1, a video model it says was sped up using findings from living neurons
30 Oct Xiaomi open-sources Xiaomi-OCR-0, a 0.8B document model it says scores 96.83 on OmniDocBench v1.6
30 Oct Tencent Cloud says AI agents now create over 20,000 TDSQL-C databases a day, up more than 80-fold in a year
29 Oct OpenAI drops a planned cut to ChatGPT Pro limits and doubles Work and Codex usage on Pro and Business Premium
29 Oct OpenAI says ChatGPT and Codex will share one token-based usage allowance from December
29 Oct Phonic launches Phonic-1, a speech-to-speech model that calls tools while the caller is still talking
28 Oct DeepSeek and Moonshot used GPT-6 Astra outputs to train new models, The Information reports
27 Oct Google DeepMind announces Genie 4, a real-time world model with native audio and text steering
26 Oct OpenAI releases GPT-6.1 Terra, a cheaper model aimed at visual reasoning and computer use
23 Oct Moonshot releases Kimi K3.1 Pro, a 2.8-trillion-parameter MoE model with a 1M-token context
22 Oct Anthropic will make auto mode the default in Claude apps for Pro, Max and Team users
21 Oct Year-old AI lab Flapping Airplanes raises a $1 billion Series A led by Index and Lux
21 Oct Anthropic previews Claude Duet, a standalone agent app that monitors connected apps and acts with approval
20 Oct Humans& launches Ampersand v0.1, a model trained to help groups pursue shared goals over time
19 Oct Google previews Agent Compute Engine microVMs for AI agents alongside a new Gemini API CLI
16 Oct Elorian releases Glimpse, a visual reasoning model that re-inspects images while it thinks
15 Oct Microsoft AI finishes pre-training MAI-Base-2, a 2.8-trillion-parameter mixture-of-experts model
15 Oct StepFun releases Step 5 Preview open weights, a 600B multimodal MoE model
14 Oct SpaceXSI releases Grok Voice 3.0, which tops Artificial Analysis's speech-to-speech index
12 Oct Meta releases Muse Wonder and Muse Wonder Full, its first frontier models for agents
9 Oct Paper finds subliminal learning can pass capabilities and backdoors between models via unrelated data
8 Oct Google Cloud previews Gemini agent coworkers with their own email, calendar and Drive
7 Oct Anthropic launches Claude Haiku 5.5, its fast low-cost model
6 Oct Google releases Nano Banana 2.1 image model with better editing and subject consistency
5 Oct Reflection AI announces Beam, a 501B open-weight mixture-of-experts model