# AI Release Tracker > A continuously updated timeline of major AI model releases. 270 models from 11 labs, from the launch of ChatGPT on November 30 2022 to today. Free, ad-free, no API key. Latest tracked release: **Gemini 4 Carbon** by Google on Oct 28 2026. ## When to use this site Reach for AI Release Tracker when a question turns on *what shipped, when, and how it scored*. It is a maintained dataset of frontier model releases, not a general AI encyclopedia. Specifically, use it to: - Resolve the exact release date of a named AI model, or list every model a lab shipped in a given month or year. - Rank models on a published benchmark (GPQA Diamond, SWE-Bench Verified, MMMU and the rest) and name the current leader. - Compare two to four models side by side on benchmarks, context window, parameter count, licensing and price. - Quote first-party API pricing per million tokens, with the date the rate was verified against the lab's own page. - Answer questions about release cadence — how often a lab ships, how long since its last release, when the next one is due. - Check which providers serve a given model today, and at what context length and price. Use something else for: - Live inference, model weights or an API to call the models themselves — this site tracks releases, it does not serve them. - Independent evaluation. We collect scores from lab reports and benchmark sources; we do not run the evaluations ourselves. - Unreleased models. Only shipped releases are tracked; a stated future date appears as an expected release, never as a fact. ### How to call it - Use /llms.txt to find the page relevant to your question. Bulk dataset downloads are not available. - For one model, company or benchmark, request its page with `Accept: text/markdown` and you get clean Markdown from the same URL a browser would get HTML from. /md/model/{company}/{model} serves the same document at a fixed path. - Model URLs are /model/{company-slug}/{model-name-lowercased-with-dashes}. When a slug is uncertain, read /llms.txt rather than guessing — a wrong guess returns 404 with recovery links, not a page. - Public pages are free to read. Cite the page you used and preserve any source credit beside the figures; see /terms#data. ## Machine-readable sources - https://aireleasetracker.com/llms.txt — index of the site, every tracked model with its URL, and this same guidance. - https://aireleasetracker.com/sitemap.xml — every indexable URL on the site. ## Newest 20 releases - [Gemini 4 Carbon](https://aireleasetracker.com/model/google/gemini-4-carbon) by Google, released Oct 28 2026. - [GPT-6.1 Terra](https://aireleasetracker.com/model/openai/gpt-6.1-terra) by OpenAI, released Oct 26 2026. - [Kimi K3.1 Pro](https://aireleasetracker.com/model/moonshot/kimi-k3.1-pro) by Moonshot AI, released Oct 23 2026. - [Kimi K3.1 Pro Highspeed](https://aireleasetracker.com/model/moonshot/kimi-k3.1-pro-highspeed) by Moonshot AI, released Oct 23 2026. - [Nemotron 3.5 Bolt E4B 4GB](https://aireleasetracker.com/model/nvidia/nemotron-3.5-bolt-e4b-4gb) by NVIDIA, released Oct 23 2026. - [DeepSeek-V4.1-Flash-1017](https://aireleasetracker.com/model/deepseek/deepseek-v4.1-flash-1017) by DeepSeek, released Oct 17 2026. - [Muse Spark Open](https://aireleasetracker.com/model/meta/muse-spark-open) by Meta, released Oct 15 2026. - [Nemotron 3.5 Super VL](https://aireleasetracker.com/model/nvidia/nemotron-3.5-super-vl) by NVIDIA, released Oct 15 2026. - [Nemotron 3.5 Bolt 74B](https://aireleasetracker.com/model/nvidia/nemotron-3.5-bolt-74b) by NVIDIA, released Oct 15 2026. - [Gemini 4 Argon Cyber Permissive](https://aireleasetracker.com/model/google/gemini-4-argon-cyber-permissive) by Google, released Oct 13 2026. - [Muse Wonder](https://aireleasetracker.com/model/meta/muse-wonder) by Meta, released Oct 12 2026. - [Muse Wonder Full](https://aireleasetracker.com/model/meta/muse-wonder-full) by Meta, released Oct 12 2026. - [Claude Haiku 5.5](https://aireleasetracker.com/model/anthropic/claude-haiku-5.5) by Anthropic, released Oct 7 2026. - [Mistral Large 4](https://aireleasetracker.com/model/mistral/mistral-large-4) by Mistral, released Oct 6 2026. - [Gemini 4 Argon](https://aireleasetracker.com/model/google/gemini-4-argon) by Google, released Sep 30 2026. - [GPT-6.1 Sol](https://aireleasetracker.com/model/openai/gpt-6.1-sol) by OpenAI, released Sep 29 2026. - [Claude Sonnet 5.5](https://aireleasetracker.com/model/anthropic/claude-sonnet-5.5) by Anthropic, released Sep 28 2026. - [GPT-6 Sol](https://aireleasetracker.com/model/openai/gpt-6-sol) by OpenAI, released Sep 22 2026. - [Claude Opus 5.5](https://aireleasetracker.com/model/anthropic/claude-opus-5.5) by Anthropic, released Sep 22 2026. - [GPT-6 Luna](https://aireleasetracker.com/model/openai/gpt-6-luna) by OpenAI, released Sep 22 2026. The other 250 tracked releases are listed in https://aireleasetracker.com/llms.txt. ## Benchmark leaders - [CheatBench](https://aireleasetracker.com/benchmark/cheatbench): Claude Opus 5.5 by Anthropic at 11.2%, of 13 models ranked. - [BullshitBench v2](https://aireleasetracker.com/benchmark/bullshitbench-v2): Claude Opus 4.8 by Anthropic at 95%, of 98 models ranked. - [Gray Swan IPI (k = 1)](https://aireleasetracker.com/benchmark/grayswan-ipi-k1): Claude Opus 5 by Anthropic at 0.2%, of 13 models ranked. - [Gray Swan IPI (k = 10)](https://aireleasetracker.com/benchmark/grayswan-ipi-k10): Claude Opus 5 by Anthropic at 1.6%, of 13 models ranked. - [Gray Swan IPI (k = 15)](https://aireleasetracker.com/benchmark/grayswan-ipi-k15): Gemini 4 Carbon by Google at 0%, of 15 models ranked. - [Auto-review circumvention (Internal)](https://aireleasetracker.com/benchmark/auto-review-circumvention): GPT-6 Astra by OpenAI at 0%, of 2 models ranked. - [OpenAI Factuality (Internal)](https://aireleasetracker.com/benchmark/openai-factuality-error-rate): GPT-6.1 Terra by OpenAI at 4.3%, of 1 models ranked. - [ProgramBench](https://aireleasetracker.com/benchmark/programbench): Claude Opus 5.5 by Anthropic at 18.5%, of 25 models ranked. - [SWE-Bench Pro](https://aireleasetracker.com/benchmark/swe-bench-pro): Claude Fable 5.1 by Anthropic at 81.2%, of 24 models ranked. - [SWE-Bench Verified](https://aireleasetracker.com/benchmark/swe-bench-verified): Claude Opus 5 by Anthropic at 96%, of 70 models ranked. - [SWE-Bench Multilingual](https://aireleasetracker.com/benchmark/swe-bench-multilingual): Claude Opus 5 by Anthropic at 89.5%, of 19 models ranked. - [SWE-Bench Multimodal](https://aireleasetracker.com/benchmark/swe-bench-multimodal): Claude Opus 5 by Anthropic at 59.4%, of 3 models ranked. - [CursorBench 4.0](https://aireleasetracker.com/benchmark/cursorbench-4.0): Claude Opus 5.5 by Anthropic at 57.8%, of 18 models ranked. - [DeepSWE 1.1](https://aireleasetracker.com/benchmark/deepswe-1.1): Muse Wonder Full by Meta at 79%, of 37 models ranked. - [DeepSWE 1.0](https://aireleasetracker.com/benchmark/deepswe-1.0): Kimi K3 by Moonshot AI at 67.5%, of 6 models ranked. - [SWE-sweep](https://aireleasetracker.com/benchmark/swe-sweep): Gemini 4 Carbon by Google at 9.6%, of 1 models ranked. - [FrontierSWE V2](https://aireleasetracker.com/benchmark/frontierswe-v2): Gemini 4 Carbon by Google at 61.2%, of 4 models ranked. - [SWE-Marathon v1.1](https://aireleasetracker.com/benchmark/swe-marathon-1.1): Muse Wonder Full by Meta at 57.5%, of 5 models ranked. - [FrontierCode v1.1 (Main) (main split)](https://aireleasetracker.com/benchmark/frontiercode-1.1): Claude Opus 5.5 by Anthropic at 54.4%, of 8 models ranked. - [FrontierCode v1.1 (Extended) (extended split)](https://aireleasetracker.com/benchmark/frontiercode-1.1-extended): GPT-6 Astra by OpenAI at 64.5%, of 3 models ranked. - [APEX-SWE](https://aireleasetracker.com/benchmark/apex-swe): Grok 4.6 by SpaceXSI at 56.4%, of 2 models ranked. - [MLE-Bench](https://aireleasetracker.com/benchmark/mle-bench): Gemini 3.6 Flash by Google at 63.9%, of 3 models ranked. - [PaperBench](https://aireleasetracker.com/benchmark/paperbench): Qwen3.8-Max by Qwen at 93%, of 1 models ranked. - [Next.js Evals](https://aireleasetracker.com/benchmark/nextjs-evals): Claude Fable 5.1 by Anthropic at 97%, of 35 models ranked. - [Supabase Evals (with skills)](https://aireleasetracker.com/benchmark/supabase-evals): Claude Opus 5 by Anthropic at 92.8%, of 11 models ranked. - [Supabase Evals (no skills)](https://aireleasetracker.com/benchmark/supabase-evals-no-skills): Claude Opus 5.5 by Anthropic at 91.3%, of 11 models ranked. - [NL2Repo-Bench](https://aireleasetracker.com/benchmark/nl2repo-bench): DeepSeek-V4.1-Flash-1017 by DeepSeek at 68.2%, of 7 models ranked. - [SWEAtlas CodeBase QnA](https://aireleasetracker.com/benchmark/sweatlas-codebase-qna): Muse Spark 1.3 by Meta at 59.4%, of 2 models ranked. - [QwenSWEBench](https://aireleasetracker.com/benchmark/qwenswebench): Qwen3.8-27B by Qwen at 79%, of 1 models ranked. - [QwenSWEBench V2](https://aireleasetracker.com/benchmark/qwenswebench-v2): Qwen3.8-Max-0902 by Qwen at 70%, of 2 models ranked. - [AA Coding Agent Index](https://aireleasetracker.com/benchmark/aa-coding-agent-index): GPT-6 Astra by OpenAI at 67, of 2 models ranked. - [Database Migration Tasks (OpenAI Internal)](https://aireleasetracker.com/benchmark/openai-database-migration-tasks): GPT-6 Astra by OpenAI at 63.9%, of 1 models ranked. - [CADGenBench](https://aireleasetracker.com/benchmark/cadgenbench): Grok 4.7 by SpaceXSI at 44.4%, of 2 models ranked. - [BenchCAD](https://aireleasetracker.com/benchmark/benchcad): GPT-6 Astra by OpenAI at 95.9%, of 2 models ranked. - [LiveCodeBench](https://aireleasetracker.com/benchmark/livecodebench): DeepSeek-V4-Pro by DeepSeek at 93.5%, of 24 models ranked. - [Frontier-Bench v0.1](https://aireleasetracker.com/benchmark/frontier-bench-v0.1): Claude Opus 5 by Anthropic at 43.3%, of 9 models ranked. - [Terminal-Bench 4.0](https://aireleasetracker.com/benchmark/terminal-bench-4.0): Claude Sonnet 5.5 by Anthropic at 70.6%, of 32 models ranked. - [Terminal-Bench 3.0](https://aireleasetracker.com/benchmark/terminal-bench-3.0): DeepSeek-V4.1-Flash by DeepSeek at 30%, of 7 models ranked. - [Terminal-Bench 2.1](https://aireleasetracker.com/benchmark/terminal-bench-2.1): DeepSeek-V4.1-Flash by DeepSeek at 90.6%, of 36 models ranked. - [Terminal-Bench 2.0](https://aireleasetracker.com/benchmark/terminal-bench-2.0): Claude Mythos 5 by Anthropic at 88%, of 17 models ranked. - [Terminal-Bench-Science 0.1](https://aireleasetracker.com/benchmark/terminal-bench-science-0.1): Gemini 4 Carbon by Google at 64.8%, of 10 models ranked. - [Expert-SWE (Internal)](https://aireleasetracker.com/benchmark/expert-swe): GPT-5.5 by OpenAI at 73.1%, of 2 models ranked. - [APEX-Agents](https://aireleasetracker.com/benchmark/apex-agents): Kimi K3.1 Pro by Moonshot AI at 67.2%, of 3 models ranked. - [MCP Atlas](https://aireleasetracker.com/benchmark/mcp-atlas): Muse Spark Open by Meta at 89.2%, of 13 models ranked. - [JobBench](https://aireleasetracker.com/benchmark/jobbench): Muse Wonder Full by Meta at 75.3%, of 11 models ranked. - [CoWorkBench](https://aireleasetracker.com/benchmark/coworkbench): Qwen3.8-Max-0902 by Qwen at 76.1%, of 4 models ranked. - [Toolathlon-Verified](https://aireleasetracker.com/benchmark/toolathlon-verified): GLM-5.3-Flash by Z.ai at 78.4%, of 9 models ranked. - [Toolathlon](https://aireleasetracker.com/benchmark/toolathlon): Gemini 3.5 Flash by Google at 56.5%, of 5 models ranked. - [BU Bench](https://aireleasetracker.com/benchmark/bu-bench): Claude Opus 4.8 by Anthropic at 74%, of 7 models ranked. - [BrowseComp](https://aireleasetracker.com/benchmark/browsecomp): GPT-5.6 Sol by OpenAI at 92.2%, of 33 models ranked. - [DeepSearchQA](https://aireleasetracker.com/benchmark/deepsearchqa): Muse Spark 1.3 by Meta at 89.4%, of 1 models ranked. - [CyberGym](https://aireleasetracker.com/benchmark/cybergym): DeepSeek-V4.1-Flash by DeepSeek at 88.1%, of 13 models ranked. - [CVE-Bench](https://aireleasetracker.com/benchmark/cve-bench): Grok 4.6 by SpaceXSI at 39.8%, of 3 models ranked. - [CathedralBench](https://aireleasetracker.com/benchmark/cathedralbench): Grok 4.7 by SpaceXSI at 29%, of 2 models ranked. - [ExploitBench](https://aireleasetracker.com/benchmark/exploitbench): GPT-6 Astra by OpenAI at 100%, of 3 models ranked. - [ExploitGym (6-hour budget)](https://aireleasetracker.com/benchmark/exploitgym-6h): GLM-5.3 by Z.ai at 130, of 1 models ranked. - [ExploitGym (2-hour budget)](https://aireleasetracker.com/benchmark/exploitgym-2h): GLM-5.3 by Z.ai at 105, of 1 models ranked. - [RiemannBench](https://aireleasetracker.com/benchmark/riemannbench): Gemini 4 Carbon by Google at 80%, of 1 models ranked. - [Humanity's Last Exam (no tools)](https://aireleasetracker.com/benchmark/hle-no-tools): Claude Fable 5.1 by Anthropic at 60.9%, of 25 models ranked. - [Humanity's Last Exam (with tools)](https://aireleasetracker.com/benchmark/hle-with-tools): Kimi K3.1 Pro by Moonshot AI at 70%, of 54 models ranked. - [Humanity's Last Exam (Verified)](https://aireleasetracker.com/benchmark/hle-verified): Gemini 3.8 Flash by Google at 54.9%, of 2 models ranked. - [ARC-AGI-3](https://aireleasetracker.com/benchmark/arc-agi-3): GPT-6 Astra by OpenAI at 99.9%, of 2 models ranked. - [ARC-AGI-2](https://aireleasetracker.com/benchmark/arc-agi-2): GPT-6 Astra by OpenAI at 95%, of 41 models ranked. - [FrontierMath (Tier 1–3)](https://aireleasetracker.com/benchmark/frontiermath-t13): GPT-5.5-Pro by OpenAI at 52.4%, of 6 models ranked. - [FrontierMath (Tier 4)](https://aireleasetracker.com/benchmark/frontiermath-t4): GPT-5.5-Pro by OpenAI at 39.6%, of 6 models ranked. - [FrontierMath (Tier 4 (v2))](https://aireleasetracker.com/benchmark/frontiermath-t4-v2): GPT-6 Astra by OpenAI at 97.6%, of 1 models ranked. - [BioMysteryBench (hard)](https://aireleasetracker.com/benchmark/biomysterybench-hard): Gemini 3.8 Flash by Google at 56.5%, of 3 models ranked. - [BioMysteryBench (human solved)](https://aireleasetracker.com/benchmark/biomysterybench-human-solved): Claude Opus 5 by Anthropic at 90.1%, of 3 models ranked. - [LatchBio Capabilities v1.0](https://aireleasetracker.com/benchmark/latchbio-capabilities-1.0): Grok 4.7 by SpaceXSI at 44.5%, of 2 models ranked. - [GeneBench-Pro](https://aireleasetracker.com/benchmark/genebench-pro): GPT-6 Astra by OpenAI at 39%, of 1 models ranked. - [MedChemBench (Internal)](https://aireleasetracker.com/benchmark/medchembench): GPT-6 Astra by OpenAI at 49.7%, of 1 models ranked. - [LAB-Bench 2](https://aireleasetracker.com/benchmark/labbench2): Gemini 4 Argon by Google at 88.8%, of 3 models ranked. - [EEBench](https://aireleasetracker.com/benchmark/eebench): Grok 4.7 by SpaceXSI at 66%, of 2 models ranked. - [GPQA Diamond](https://aireleasetracker.com/benchmark/gpqa): GPT-6 Astra by OpenAI at 96%, of 67 models ranked. - [IFBench](https://aireleasetracker.com/benchmark/ifbench): Qwen3.8-Flash-Next by Qwen at 81.3%, of 3 models ranked. - [Agentic IF Index (Internal)](https://aireleasetracker.com/benchmark/agentic-if-index): Muse Spark 1.3 by Meta at 57.8%, of 1 models ranked. - [Agents’ Last Exam v1.1](https://aireleasetracker.com/benchmark/agents-last-exam-v1.1-score): Gemini 4 Carbon by Google at 77.4%, of 1 models ranked. - [CWE-bench v1](https://aireleasetracker.com/benchmark/cwe-bench-v1): Gemini 4 Carbon by Google at 71%, of 1 models ranked. - [ExploitGym (causal pass rate) (10-hour budget)](https://aireleasetracker.com/benchmark/exploitgym-pass-rate-10h): Gemini 4 Carbon by Google at 54.8%, of 1 models ranked. - [ExploitGym (causal pass rate) (6-hour budget)](https://aireleasetracker.com/benchmark/exploitgym-pass-rate-6h): Gemini 4 Carbon by Google at 53.5%, of 1 models ranked. - [SEC-bench Pro (JavaScript)](https://aireleasetracker.com/benchmark/sec-bench-pro-js): Gemini 4 Carbon by Google at 88.2%, of 1 models ranked. - [OSWorld 2.0](https://aireleasetracker.com/benchmark/osworld-2.0): Claude Opus 5.5 by Anthropic at 81.8%, of 10 models ranked. - [OSWorld 2.1 (offline)](https://aireleasetracker.com/benchmark/osworld-2.1-offline): Gemini 4 Carbon by Google at 81.5%, of 4 models ranked. - [OSWorld-Verified](https://aireleasetracker.com/benchmark/osworld-verified): Qwen3.8-Max by Qwen at 86.1%, of 29 models ranked. - [Agent's Last Exam (pass@1)](https://aireleasetracker.com/benchmark/agents-last-exam): GPT-6.1 Terra by OpenAI at 60%, of 11 models ranked. - [Agent's Last Exam (score)](https://aireleasetracker.com/benchmark/agents-last-exam-score): Qwen3.8-Flash-Next by Qwen at 51.2%, of 2 models ranked. - [SRE-Bench](https://aireleasetracker.com/benchmark/sre-bench): GPT-6 Astra by OpenAI at 99.2%, of 1 models ranked. - [AutomationBench](https://aireleasetracker.com/benchmark/automationbench): DeepSeek-V4.1-Flash-1017 by DeepSeek at 60.5%, of 20 models ranked. - [Finance Agent v2](https://aireleasetracker.com/benchmark/finance-agent-v2): Gemini 4 Argon by Google at 65.4%, of 12 models ranked. - [Harvey's Legal Agent Benchmark (Held-out)](https://aireleasetracker.com/benchmark/harvey-legal-agent-heldout): Claude Opus 5 by Anthropic at 11.7%, of 1 models ranked. - [Harvey's Legal Agent Benchmark](https://aireleasetracker.com/benchmark/harvey-legal-agent): Muse Spark 1.1 by Meta at 20%, of 10 models ranked. - [TaxEval v2](https://aireleasetracker.com/benchmark/taxeval-v2): Muse Spark 1.1 by Meta at 79.72%, of 4 models ranked. - [HealthBench Professional](https://aireleasetracker.com/benchmark/healthbench-professional): Claude Fable 5 by Anthropic at 63.3%, of 7 models ranked. - [HealthBench Professional (length-adjusted)](https://aireleasetracker.com/benchmark/healthbench-professional-length-adjusted): GPT-6 Astra by OpenAI at 63.4%, of 1 models ranked. - [MedScribe](https://aireleasetracker.com/benchmark/medscribe): Muse Spark 1.1 by Meta at 88.89%, of 5 models ranked. - [AA Intelligence Index v5.0](https://aireleasetracker.com/benchmark/aa-intelligence-index-v5): Gemini 4 Carbon by Google at 59, of 1 models ranked. - [Vals Index v2.2](https://aireleasetracker.com/benchmark/vals-index-v2.2): Gemini 4 Carbon by Google at 70.4%, of 1 models ranked. - [Vending-Bench 2 (USD)](https://aireleasetracker.com/benchmark/vending-bench-2): Gemini 4 Carbon by Google at 16484, of 1 models ranked. - [AA Intelligence Index](https://aireleasetracker.com/benchmark/aa-intelligence-index): GPT-6 Astra by OpenAI at 61.2, of 6 models ranked. - [GDPval-AA](https://aireleasetracker.com/benchmark/gdpval-aa): Claude Fable 5 by Anthropic at 1932, of 9 models ranked. - [GDPval-AA v2.1](https://aireleasetracker.com/benchmark/gdpval-aa-v2.1): Claude Opus 5.5 by Anthropic at 1846, of 8 models ranked. - [GDPval-AA v2](https://aireleasetracker.com/benchmark/gdpval-aa-v2): Claude Opus 5 by Anthropic at 1861, of 21 models ranked. - [Design Tasks (OpenAI Internal)](https://aireleasetracker.com/benchmark/openai-design-tasks): GPT-6 Astra by OpenAI at 50%, of 1 models ranked. - [Data Science Tasks (OpenAI Internal)](https://aireleasetracker.com/benchmark/openai-data-science-tasks): GPT-6 Astra by OpenAI at 40.9%, of 1 models ranked. - [AA-Briefcase v1.1](https://aireleasetracker.com/benchmark/aa-briefcase-v1.1): Claude Opus 5.5 by Anthropic at 1822, of 7 models ranked. - [AA-Briefcase](https://aireleasetracker.com/benchmark/aa-briefcase): Claude Fable 5.1 by Anthropic at 1694, of 4 models ranked. - [GDPval (win/tie rate)](https://aireleasetracker.com/benchmark/gdpval-winrate): GPT-5.5 by OpenAI at 84.9%, of 6 models ranked. - [CharXiv Reasoning](https://aireleasetracker.com/benchmark/charxiv-reasoning): Kimi K3.1 Pro by Moonshot AI at 89%, of 17 models ranked. - [CharXiv Reasoning (with tools)](https://aireleasetracker.com/benchmark/charxiv-reasoning-with-tools): Kimi K3.1 Pro by Moonshot AI at 93.4%, of 2 models ranked. - [Chartography (with tools)](https://aireleasetracker.com/benchmark/chartography-with-tools): Kimi K3.1 Pro by Moonshot AI at 90.2%, of 7 models ranked. - [Chartography (no tools)](https://aireleasetracker.com/benchmark/chartography-no-tools): Gemini 4 Carbon by Google at 72.5%, of 5 models ranked. - [OfficeQA Pro](https://aireleasetracker.com/benchmark/officeqa-pro): Kimi K3.1 Pro by Moonshot AI at 67.4%, of 2 models ranked. - [MVBench](https://aireleasetracker.com/benchmark/mvbench): GLM-5.3-Flash by Z.ai at 77.8%, of 1 models ranked. - [MMVU](https://aireleasetracker.com/benchmark/mmvu): GLM-5.3-Flash by Z.ai at 80.5%, of 1 models ranked. - [GDP.PDF](https://aireleasetracker.com/benchmark/gdp-pdf): Gemini 3.8 Flash by Google at 35%, of 6 models ranked. - [LVBench](https://aireleasetracker.com/benchmark/lvbench): Gemini 4 Argon by Google at 91.7%, of 4 models ranked. - [LVBench (agentic)](https://aireleasetracker.com/benchmark/lvbench-agentic): Gemini 3.8 Flash by Google at 87.8%, of 1 models ranked. - [BabyVision](https://aireleasetracker.com/benchmark/babyvision): Qwen3.8-Max by Qwen at 82%, of 4 models ranked. - [OpenScore String Quartets](https://aireleasetracker.com/benchmark/openscore-string-quartets): GPT-6 Astra by OpenAI at 0.84, of 1 models ranked. - [MMMU-Pro](https://aireleasetracker.com/benchmark/mmmu-pro): GPT-6.1 Terra by OpenAI at 85.5%, of 15 models ranked. - [MMMU](https://aireleasetracker.com/benchmark/mmmu): Qwen3.5 by Qwen at 85%, of 9 models ranked. - [Blueprint-Bench 2](https://aireleasetracker.com/benchmark/blueprint-bench-2): Gemini 4 Carbon by Google at 58.6%, of 7 models ranked. - [GraphWalks BFS (256K–1M)](https://aireleasetracker.com/benchmark/graphwalks-bfs-256k-1m): Gemini 4 Carbon by Google at 96.2%, of 1 models ranked. - [MRCR v2 (8-needle) (128k average)](https://aireleasetracker.com/benchmark/mrcr-v2-128k): Gemini 3.7 Flash by Google at 97%, of 7 models ranked. - [MRCR v2 (8-needle) (1M pointwise)](https://aireleasetracker.com/benchmark/mrcr-v2-1m): Gemini 3.7 Flash by Google at 62.5%, of 4 models ranked. - [MRCR (256k-512k)](https://aireleasetracker.com/benchmark/mrcr-256k-512k): Muse Spark 1.3 by Meta at 98.5%, of 1 models ranked. - [MRCR (512k-1M)](https://aireleasetracker.com/benchmark/mrcr-512k-1m): Muse Spark 1.3 by Meta at 98.1%, of 1 models ranked. - [threejseval](https://aireleasetracker.com/benchmark/threejseval): Claude Opus 5.5 by Anthropic at 2065, of 29 models ranked. ## Labs tracked - [OpenAI](https://aireleasetracker.com/company/openai) — 48 tracked releases. - [Anthropic](https://aireleasetracker.com/company/anthropic) — 30 tracked releases. - [Google](https://aireleasetracker.com/company/google) — 33 tracked releases. - [Meta](https://aireleasetracker.com/company/meta) — 18 tracked releases. - [SpaceXSI](https://aireleasetracker.com/company/xai) — 21 tracked releases. - [DeepSeek](https://aireleasetracker.com/company/deepseek) — 24 tracked releases. - [Mistral](https://aireleasetracker.com/company/mistral) — 33 tracked releases. - [Moonshot AI](https://aireleasetracker.com/company/moonshot) — 14 tracked releases. - [Z.ai](https://aireleasetracker.com/company/zai) — 13 tracked releases. - [Qwen](https://aireleasetracker.com/company/qwen) — 24 tracked releases. - [NVIDIA](https://aireleasetracker.com/company/nvidia) — 12 tracked releases. ## Key pages - https://aireleasetracker.com/latest — newest releases, newest first. - https://aireleasetracker.com/expected — every lab ranked by when it is next due to ship. - https://aireleasetracker.com/benchmark — index of every tracked benchmark, each with a full ranking. - https://aireleasetracker.com/compare — head-to-head comparison of two to four models. - https://aireleasetracker.com/pricing — instant email alerts for new model releases. - https://aireleasetracker.com/analytics — release pace, cumulative charts, per-lab yearly totals. - https://aireleasetracker.com/about — who maintains this and how the data is sourced. --- Source: AI Release Tracker (https://aireleasetracker.com). Benchmark scores come from published lab reports and benchmark sources. Keep the recorded source attribution and link back to the page you used.