AINews Portal

🏆 AI Benchmarks Index

What each major benchmark measures and who leads it — every score served from our own tracking pipeline and refreshed daily. JSON is available at /api/benchmarks.

🔬 GPQA Diamond

Graduate-level science questions designed to be Google-proof — tests deep reasoning.

  1. 1GPT-6.2 Astra96.8%
  2. 2Gemini 4 Hydrogen96.8%
  3. 3Gemini 4 Carbon96.5%
  4. 4Gemini 4 Argon96.4%
  5. 5Claude Fable 5.296.0%
  6. 6GPT-6 Astra95.8%
  7. 7Claude Sonnet 5.595.6%
  8. 8GPT-6.1 Sol95.4%
  9. 9Gemini 3.8 Flash95.4%
  10. 10Gemini 3.7 Flash94.8%
💻 SWE-Bench Verified

Real GitHub issues an agent must fix end-to-end — the standard for coding agents.

  1. 1Claude Opus 4.7 (Fast)83.5%
  2. 2Claude Opus 4.783.5%
  3. 3GPT-5.580.6%
  4. 4Gemini 3.5 Flash79.3%
  5. 5Claude Opus 4.678.7%
  6. 6GLM 5.278.7%
  7. 7DeepSeek V4 Pro77.6%
  8. 8Qwen3.7 Max77.3%
  9. 9GPT-5.476.9%
  10. 10Claude Opus 4.576.7%
➗ MATH Level 5

The hardest tier of competition math problems from the MATH benchmark.

  1. 1GPT-598.1%
  2. 2GPT-5 Mini97.8%
  3. 3o4 Mini97.8%
  4. 4o397.8%
  5. 5Claude Sonnet 4.597.7%
  6. 6Qwen3 Max97.1%
  7. 7R196.6%
  8. 8o3 Mini96.5%
  9. 9Claude Haiku 4.596.4%
  10. 10Gemini 2.5 Pro95.9%
🧮 OTIS Mock AIME

Olympiad-style mock AIME exam (2024–25) — elite high-school competition math.

  1. 1Qwen3.8 Max (0902)100.0%
  2. 2Claude Opus 5.5100.0%
  3. 3GPT-6.2 Astra100.0%
  4. 4GPT-5.6 Sol100.0%
  5. 5GPT-5.5 Pro100.0%
  6. 6Gemini 4 Hydrogen100.0%
  7. 7Claude Fable 5100.0%
  8. 8Gemini 4 Carbon100.0%
  9. 9GPT-5.5100.0%
  10. 10Claude Sonnet 5.5100.0%
♾️ FrontierMath

Research-level math problems written by mathematicians — the hardest math benchmark.

  1. 1GPT-5.5 Pro52.4%
  2. 2GPT-5.551.7%
  3. 3GPT-5.4 Pro50.0%
  4. 4GPT-5.447.6%
  5. 5Claude Opus 4.8 (Fast)47.2%
  6. 6Claude Opus 4.847.2%
  7. 7Claude Opus 4.743.8%
  8. 8Claude Opus 4.7 (Fast)43.8%
  9. 9GPT-5.240.7%
  10. 10Claude Opus 4.640.7%
✅ SimpleQA Verified

Short factual questions measuring hallucination resistance — can it just be right?

  1. 1Claude Fable 5.294.0%
  2. 2GPT-6.2 Astra78.5%
  3. 3Gemini 4 Carbon77.7%
  4. 4Gemini 4 Argon76.0%
  5. 5GPT-6 Astra75.6%
  6. 6GPT-6.1 Sol73.9%
  7. 7Claude Opus 5.572.2%
  8. 8Gemini 4 Hydrogen70.9%
  9. 9Claude Fable 5.170.8%
  10. 10Claude Fable 570.7%
💬 LMArena (Text)

Human preference Elo from blind head-to-head chat battles — the de-facto vibes benchmark.

  1. 1claude-opus-5-high1502
  2. 2Claude Opus 4.61497
  3. 3Claude Fable 51493
  4. 4MiMo-V2.6-Pro1492
  5. 5claude-fable-5-high1492
  6. 6Qwen3.8 Max1486
  7. 7Claude Opus 4.71483
  8. 8Gemini 3.5 Flash1482
  9. 9Gemini 3.1 Pro Preview1480
  10. 10Muse Spark 1.11479
🖼️ LMArena (Image)

Human preference Elo for text-to-image generation battles.

  1. 1gpt-image-2.5-sunburst1429
  2. 2GPT-Image-2.6 Sunburst (Medium)1410
  3. 3gpt-image-2.5-flare1400
  4. 4MAI-Image-2.71387
  5. 5gpt-image-2 (medium)1382
  6. 6GPT-Image-2.6 Flare (Medium)1375
  7. 7grok-imagine-image-2.0 (canvas)1334
  8. 8mai-image-2.61331
  9. 9Nano Banana 2.11328
  10. 10Grok Imagine Image 2.01318
🎬 LMArena (Video)

Human preference Elo for image-to-video generation battles.

  1. 1Gemini Omni 1.2 Flash1503
  2. 2dreamina-seedance-2.5-720p1494
  3. 3minimax-h31493
  4. 4gemini-omni-1.1-flash1487
  5. 5wan3.01482
  6. 6dreamina-seedance-2.0-720p1480
  7. 7SkyReels V41475
  8. 8grok-imagine-video-1.5-preview-720p1466
  9. 9gemini-omni-flash1463
  10. 10hidream-o1-video-1.01458

Score data: Epoch AI Benchmarking Hub (CC-BY) and LMArena (CC-BY), ingested daily and matched to our model catalog. Rankings are never for sale — see the methodology and full model leaderboard.