OpenAI drops another batch of mathematical breakthroughs
OpenAI reports an unreleased frontier model produced solutions to long-standing math problems across 722 manuscripts, suggesting a major leap in AI mathematical reasoning.
The latest model from every major lab, what it costs to use, how often each lab ships, and the breakthroughs worth knowing about, in one place.
The moments that move the field: new frontier models, open problems solved, landmark rulings.
OpenAI reports an unreleased frontier model produced solutions to long-standing math problems across 722 manuscripts, suggesting a major leap in AI mathematical reasoning.
Mistral released Mistral Large 4, a new large multimodal model positioned to outperform both closed and open rivals.
Mistral released a new trillion-parameter open-weight model, positioning it as a frontier-level competitor outside China.
NVIDIA claims GPT-6 Astra Ultrafast is available in the OpenAI API and to some ChatGPT/Codex users, accelerated on Blackwell GPUs.
The Verge reports OpenAI announced “Dots,” an AI agent powered by GPT-6 at DevDay, signaling a major new model/agent push against Meta.
Google released a new Gemini model it calls its most powerful yet, positioned as a workhorse for coding and cybersecurity use cases.
Google announced Gemini 4 Argon as its next AI frontier model and is restricting access to trusted cyber defenders, indicating high capability and risk.
Ars Technica reports Google announced a new Gemini 4 Argon model, indicating a new major model generation even if not yet available.
Newest release first. Prices are US dollars per million tokens of input and output, the way the labs bill API use. "Typical gap" is the median time between release days over the past year.
| Lab | Latest model | Released | Input | Output | Context | Last 90 days | Typical gap |
|---|---|---|---|---|---|---|---|
| Anthropic | Claude Haiku 5.5 Also: Claude Sonnet 5.5 (Sep 28), Claude Opus 5.5 (Sep 22) | todayOct 07, 2026 | $0.10 | $0.50 | 1M | 5 models | 22 days |
| Nano Banana 2.1 Also: Gemini 3.8 Flash (Sep 02), Gemini 3.7 Flash (Aug 13) | yesterdayOct 06, 2026 | $1.50 | $7.50 | 66K | 5 models | 20 days | |
| Mistral | Mistral Large 4 Also: Mistral Medium 3.5 (Apr 30), Mistral Small 4 (Mar 16) | yesterdayOct 06, 2026 | $0.68 | $2.09 | 524K | 1 model | 38 days |
| OpenAI | GPT-6.1 Sol Pro Also: GPT-6.1 Sol (Sep 29), GPT-6 Luna Pro (Sep 22) | 8 days agoSep 29, 2026 | $2.00 | $10.00 | 1.05M | 8 models | 15 days |
| Z.ai (GLM) | GLM 5.3 Prime Also: GLM 5.3 FlashX (Sep 18), GLM 5.3 Flash (Aug 26) | 14 days agoSep 23, 2026 | $2.80 | $8.80 | 1M | 4 models | 23 days |
| Alibaba (Qwen) | Qwen3.8 Max Prime Also: Qwen3.8 Omni Flash (Sep 21), Qwen3.8 Max (0902) (Sep 03) | 14 days agoSep 23, 2026 | $4.00 | $12.00 | 1M | 7 models | 13 days |
| Cohere | Command A+ Also: Command A (Mar 13), Command R7B (12-2024) (Dec 14) | 15 days agoSep 22, 2026 | $0.30 | $1.50 | 192K | 1 model | — |
| NVIDIA | Switchyard Also: Nemotron 3.5 Lightning (Aug 11), Nemotron 3.5 Content Safety (Jun 04) | 16 days agoSep 21, 2026 | — | — | 1M | 2 models | 76 days |
| Xiaomi | MiMo-V2.6-Pro-UltraSpeed Also: MiMo-V2.6-Flash (Sep 21), MiMo-V2.6-Pro (Sep 21) | 16 days agoSep 21, 2026 | $4.35 | $8.70 | 1.05M | 3 models | — |
| xAI | Grok 4.7 Also: Grok 4.6 (Aug 12), Grok 4.5 (Jul 08) | 16 days agoSep 21, 2026 | $2.00 | $6.00 | 500K | 2 models | 35 days |
| DeepSeek | DeepSeek V4.1 Flash Also: DeepSeek V4 Flash Vision Exp (Aug 21), DeepSeek V4 Pro 0813 (Aug 12) | 27 days agoSep 10, 2026 | $0.30 | $1.20 | 1.05M | 4 models | 20 days |
| Meta | Muse Spark 1.3 Contributor Also: Muse Spark 1.3 (Sep 02), Muse Spark 1.2 Contributor (Aug 21) | 35 days agoSep 02, 2026 | $0.10 | $0.20 | 1.05M | 6 models | 12 days |
| Tencent | Hy4 preview Also: Hy-MT2-1.8B (Aug 20), Hy-MT2-30B-A3B (Aug 20) | 40 days agoAug 28, 2026 | $0.75 | $2.25 | 1.05M | 4 models | 26 days |
| ByteDance Seed | Seed 2.1 Turbo Also: Seed-2.0-Code (Aug 12), Seed-2.0-Lite (Mar 10) | 56 days agoAug 12, 2026 | $0.50 | $2.50 | 262K | 2 models | 65 days |
| Thinking Machines | Inkling Small Also: Inkling (Jul 17) | 69 days agoJul 30, 2026 | $0.45 | $1.20 | 524K | 2 models | — |
| Moonshot AI (Kimi) | Kimi K3 Also: Kimi K2.7 Code (Jun 12), Kimi K2.6 (Apr 20) | 83 days agoJul 16, 2026 | $0.50 | $15.00 | 1.05M | 1 model | 68 days |
| MiniMax | MiniMax M3 Also: MiniMax M2.7 (Mar 18), MiniMax M2.5 (Feb 12) | 129 days agoMay 31, 2026 | $0.30 | $1.20 | 1.05M | 0 models | 34 days |
| Amazon | Nova 2 Lite Also: Nova Premier 1.0 (Oct 31), Nova Lite 1.0 (Dec 05) | 309 days agoDec 02, 2025 | $0.30 | $2.50 | 1M | 0 models | — |
| Perplexity | Sonar Pro Search Also: Sonar Reasoning Pro (Mar 07), Sonar Pro (Mar 07) | 342 days agoOct 30, 2025 | $3.00 | $15.00 | 200K | 0 models | — |
| Microsoft | Phi 4 Also: WizardLM-2 8x22B (Apr 16) | 635 days agoJan 10, 2025 | $0.07 | $0.14 | 16K | 0 models | — |
New models per week over the last six months, by lab.
Models whose price changed in the last 30 days, per million tokens.
| Model | Lab | Input | Output | Changed |
|---|---|---|---|---|
| DeepSeek V4.1 Flash | DeepSeek | $0.05 → $0.30 | $1.20 → $1.20 | Oct 07 |
| Hy4 preview | Tencent | $0.83 → $0.75 | $2.50 → $2.25 | Oct 07 |
| Kimi K3 | Moonshot AI (Kimi) | $0.62 → $0.50 | $15.00 → $15.00 | Oct 07 |
| Gemma 4 26B A4B | $0.09 → $0.08 | $0.30 → $0.26 | Oct 07 | |
| Hy3 | Tencent | $0.13 → $0.08 | $0.53 → $0.33 | Oct 07 |
| Nemotron 3.5 Lightning | NVIDIA | $0.06 → $0.05 | $0.16 → $0.14 | Oct 07 |
| DeepSeek V4 Pro 0813 | DeepSeek | $1.32 → $0.66 | $3.96 → $1.98 | Oct 07 |
| GLM 5.2 | Z.ai (GLM) | $0.03 → $0.17 | $12.00 → $7.20 | Oct 07 |
| DeepSeek V4 Flash 0731 | DeepSeek | $0.02 → $0.02 | $1.28 → $1.28 | Oct 07 |
| Kimi K2.6 | Moonshot AI (Kimi) | $0.95 → $0.47 | $4.00 → $2.45 | Oct 07 |
The last two weeks from the labs' own blogs, tech outlets' AI desks and research publications.
Share of the headlines Web Pulse reads on news front pages each day that are about AI.
Search every one in the Headline Archive.
Models and prices come from OpenRouter's public catalogue, which serves most labs' models through one API. The release date is the day a model was listed there, usually its launch day. Prices are the lab's standard API rate; free, batch and preview variants are left out. Last checked Oct 07, 19:30 UTC.
News comes from OpenAI, Google DeepMind, Google AI, Hugging Face, NVIDIA, Ars Technica, TechCrunch, The Verge, Wired, MIT Technology Review, Quanta Magazine, Nature, read every couple of hours. An AI model sorts each story by topic and flags milestones; it only sees each story's title and summary. Anthropic and Meta don't publish news feeds, so their releases show up through the model catalogue and press coverage.
Pulse Index watches YouTube, RSS feeds, websites and news front pages for your keywords and tells you when they show up.