# tensor.news

> The evidence layer for AI model claims: benchmark leaderboards with integrity grades, a model claim ledger, and head-to-head comparisons.

## Core surfaces

- [Benchmark Integrity Index](https://tensor.news/benchmarks): every tracked benchmark graded 0-100 on discrimination, saturation, contamination, and harness comparability.
- [Models](https://tensor.news/models): every tracked model with independent benchmark scores and Model Claim Ledger status.
- [Compare](https://tensor.news/compare): curated head-to-head model comparisons.
- [Labs](https://tensor.news/labs): organizations building the tracked models.
- [Papers](https://tensor.news/papers): research papers behind the tracked models.
- [Methodology](https://tensor.news/methodology): how the Integrity score and Claim Ledger work.
- [About](https://tensor.news/about)

## Benchmarks

- [APEX-Agents](https://tensor.news/benchmarks/apex-agents): integrity 100/100 (A)
- [Chess Puzzles](https://tensor.news/benchmarks/chess-puzzles): integrity 100/100 (A)
- [HLE](https://tensor.news/benchmarks/hle): integrity 100/100 (A)
- [SimpleQA Verified](https://tensor.news/benchmarks/simpleqa-verified): integrity 100/100 (A)
- [ARC-AGI-2](https://tensor.news/benchmarks/arc-agi-2): integrity 99/100 (A)
- [FrontierMath-2025-02-28-Private](https://tensor.news/benchmarks/frontiermath-2025-02-28-private): integrity 99/100 (A)
- [FrontierMath-Tier-4-v2-Private](https://tensor.news/benchmarks/frontiermath-tier-4-v2-private): integrity 99/100 (A)
- [CadEval](https://tensor.news/benchmarks/cadeval): integrity 98/100 (A)
- [CursorBench](https://tensor.news/benchmarks/cursorbench): integrity 98/100 (A)
- [FrontierMath-Tiers-1-3-v2-Private](https://tensor.news/benchmarks/frontiermath-tiers-1-3-v2-private): integrity 98/100 (A)
- [GBAEval](https://tensor.news/benchmarks/gbaeval): integrity 98/100 (A)
- [GDPval](https://tensor.news/benchmarks/gdpval): integrity 98/100 (A)
- [FrontierMath-Tier-4-2025-07-01-Private](https://tensor.news/benchmarks/frontiermath-tier-4-2025-07-01-private): integrity 93/100 (A)
- [CritPt](https://tensor.news/benchmarks/critpt): integrity 92/100 (A)
- [The Agent Company](https://tensor.news/benchmarks/the-agent-company): integrity 92/100 (A)
- [GSO-Bench](https://tensor.news/benchmarks/gso-bench): integrity 90/100 (A)
- [Balrog](https://tensor.news/benchmarks/balrog): integrity 89/100 (A)
- [ExploitBench](https://tensor.news/benchmarks/exploitbench): integrity 89/100 (A)
- [GeoBench](https://tensor.news/benchmarks/geobench): integrity 89/100 (A)
- [PostTrainBench](https://tensor.news/benchmarks/posttrainbench): integrity 89/100 (A)
- [OTIS Mock AIME 2024-2025](https://tensor.news/benchmarks/otis-mock-aime-2024-2025): integrity 88/100 (A)
- [WeirdML](https://tensor.news/benchmarks/weirdml): integrity 88/100 (A)
- [Aider polyglot](https://tensor.news/benchmarks/aider-polyglot): integrity 87/100 (A)
- [Fiction.LiveBench](https://tensor.news/benchmarks/fiction-livebench): integrity 87/100 (A)
- [VPCT](https://tensor.news/benchmarks/vpct): integrity 87/100 (A)
- [SimpleBench](https://tensor.news/benchmarks/simplebench): integrity 86/100 (A)
- [Terminal Bench](https://tensor.news/benchmarks/terminal-bench): integrity 84/100 (B)
- [SWE-Bench verified](https://tensor.news/benchmarks/swe-bench-verified): integrity 83/100 (B)
- [CL-bench](https://tensor.news/benchmarks/cl-bench): integrity 82/100 (B)
- [CL-bench Life](https://tensor.news/benchmarks/cl-bench-life): integrity 81/100 (B)
- [GPQA diamond](https://tensor.news/benchmarks/gpqa-diamond): integrity 81/100 (B)
- [OSWorld 2.0](https://tensor.news/benchmarks/osworld-2-0): integrity 81/100 (B)
- [Remote Labor Index](https://tensor.news/benchmarks/remote-labor-index): integrity 81/100 (B)
- [OSWorld](https://tensor.news/benchmarks/osworld): integrity 80/100 (B)
- [Cybench](https://tensor.news/benchmarks/cybench): integrity 79/100 (B)
- [DeepResearch Bench](https://tensor.news/benchmarks/deepresearch-bench): integrity 78/100 (B)
- [Lech Mazur Writing](https://tensor.news/benchmarks/lech-mazur-writing): integrity 76/100 (B)
- [ARC-AGI](https://tensor.news/benchmarks/arc-agi): integrity 74/100 (B)
- [MATH level 5](https://tensor.news/benchmarks/math-level-5): integrity 73/100 (B)
- [BBH](https://tensor.news/benchmarks/bbh): integrity 71/100 (B)
- [GSM8K](https://tensor.news/benchmarks/gsm8k): integrity 69/100 (C)
- [ScienceQA](https://tensor.news/benchmarks/scienceqa): integrity 68/100 (C)
- [HellaSwag](https://tensor.news/benchmarks/hellaswag): integrity 67/100 (C)
- [MMLU](https://tensor.news/benchmarks/mmlu): integrity 67/100 (C)
- [ARC AI2](https://tensor.news/benchmarks/arc-ai2): integrity 66/100 (C)
- [TriviaQA](https://tensor.news/benchmarks/triviaqa): integrity 63/100 (C)
- [PIQA](https://tensor.news/benchmarks/piqa): integrity 62/100 (C)
- [Winogrande](https://tensor.news/benchmarks/winogrande): integrity 62/100 (C)
- [OpenBookQA](https://tensor.news/benchmarks/openbookqa): integrity 61/100 (C)
- [ANLI](https://tensor.news/benchmarks/anli): integrity 59/100 (C)
- [LAMBADA](https://tensor.news/benchmarks/lambada): integrity 54/100 (D)

## Top models

- [GPT-5.5](https://tensor.news/models/gpt-5-5): best score 100
- [GPT-5.5 Pro](https://tensor.news/models/gpt-5-5-pro): best score 100
- [Claude Fable 5](https://tensor.news/models/claude-fable-5): best score 99.72
- [Claude Opus 4.8](https://tensor.news/models/claude-opus-4-8): best score 98.33
- [GPT-5](https://tensor.news/models/gpt-5): best score 98.13
- [Gemini 3.1 Pro](https://tensor.news/models/gemini-3-1-pro): best score 98
- [GPT-5 mini](https://tensor.news/models/gpt-5-mini): best score 97.85
- [o4-mini](https://tensor.news/models/o4-mini): best score 97.83
- [Claude Opus 4.7](https://tensor.news/models/claude-opus-4-7): best score 97.8
- [o3](https://tensor.news/models/o3): best score 97.77
- [Claude Sonnet 4.5](https://tensor.news/models/claude-sonnet-4-5): best score 97.73
- [o3-pro](https://tensor.news/models/o3-pro): best score 97.2
- [Qwen3-Max](https://tensor.news/models/qwen3-max): best score 97.13
- [DeepSeek-V4-Pro](https://tensor.news/models/deepseek-v4-pro): best score 96.66
- [DeepSeek-R1 (May 2025)](https://tensor.news/models/deepseek-r1-may-2025): best score 96.64
- [o3-mini](https://tensor.news/models/o3-mini): best score 96.49
- [Kimi K2.7 Code](https://tensor.news/models/kimi-k2-7-code): best score 96.39
- [Claude Haiku 4.5](https://tensor.news/models/claude-haiku-4-5): best score 96.36
- [Kimi K2.6](https://tensor.news/models/kimi-k2-6): best score 96.11
- [GPT-5.2](https://tensor.news/models/gpt-5-2): best score 96.11
- [Gemini 2.5 Pro (May 2025)](https://tensor.news/models/gemini-2-5-pro-may-2025): best score 95.9
- [Gemini 2.5 Pro (Mar 2025)](https://tensor.news/models/gemini-2-5-pro-mar-2025): best score 95.56
- [Gemini 3.5 Flash](https://tensor.news/models/gemini-3-5-flash): best score 95.55
- [GPT-5.4](https://tensor.news/models/gpt-5-4): best score 95.3
- [GPT-5 nano](https://tensor.news/models/gpt-5-nano): best score 95.24
- [Qwen3.7-Max](https://tensor.news/models/qwen3-7-max): best score 94.99
- [o1](https://tensor.news/models/o1): best score 94.71
- [GPT-5.4 Pro](https://tensor.news/models/gpt-5-4-pro): best score 94.5
- [Claude Opus 4.6](https://tensor.news/models/claude-opus-4-6): best score 94.44
- [Grok 4](https://tensor.news/models/grok-4): best score 94.4

## Labs

- [Qwen](https://tensor.news/labs/qwen)
- [Meta AI](https://tensor.news/labs/meta-ai)
- [OpenAI](https://tensor.news/labs/openai)
- [DeepSeek](https://tensor.news/labs/deepseek)
- [NVIDIA](https://tensor.news/labs/nvidia)
- [Google DeepMind](https://tensor.news/labs/google-deepmind)
- [EleutherAI](https://tensor.news/labs/eleutherai)
- [Zhipu AI](https://tensor.news/labs/zhipu-ai)
- [Hugging Face](https://tensor.news/labs/hugging-face)
- [IBM Research](https://tensor.news/labs/ibm-research)
- [Red Hat AI](https://tensor.news/labs/red-hat-ai)
- [Apple](https://tensor.news/labs/apple)
- [H2O.ai](https://tensor.news/labs/h2o-ai)
- [MiniMax](https://tensor.news/labs/minimax)
- [Moonshot AI](https://tensor.news/labs/moonshot-ai)
- [Mistral AI](https://tensor.news/labs/mistral-ai)
- [Microsoft](https://tensor.news/labs/microsoft)
- [Technology Innovation Institute](https://tensor.news/labs/tii)
- [Nous Research](https://tensor.news/labs/nous-research)

## Notes

All entity pages carry JSON-LD structured data (SoftwareApplication for models, Dataset for benchmarks, Organization for labs, ScholarlyArticle for papers) plus a BreadcrumbList. A fuller machine-readable digest with leaderboards is available at [/llms-full.txt](https://tensor.news/llms-full.txt).