MMMU
MMMU tests multimodal reasoning at college-exam level — image + text questions across 30 subjects from art history to medical diagnostics. Gemini 3 Pro leads at ~84%, with GPT-5.5 and Claude 4.7 in the high 70s. The cleanest signal for "is this model genuinely multimodal or just text-with-vision-grafted-on".
Category: multimodal · Source: mmmu-benchmark.github.io ↗
| # | Model | Score | Setting | Measured | Source |
|---|---|---|---|---|---|
| 01 |  Gemini 3.5 | 86.0% | val | May 20, 2026 | link ↗ |
| 02 | Gemini 3 Pro | 84.3% | val | Apr 22, 2026 | link ↗ |
| 03 | Gemini Omni | 82.5% | val | May 20, 2026 | link ↗ |
| 04 |  Claude 4.7 OpusAnthropic | 79.0% | val | Apr 16, 2026 | link ↗ |
| 05 |  GPT-5.5OpenAI | 78.4% | val | Apr 23, 2026 | link ↗ |
| 06 | GPT-5 OpenAI | 74.1% | val | Aug 7, 2025 | link ↗ |
| 07 | Gemini 2.5 Pro | 72.4% | val | Jun 17, 2025 | link ↗ |
| 08 |  Qwen3.6-27BAlibaba | 70.3% | val | Apr 22, 2026 | link ↗ |
| 09 |  Llama 4 MaverickMeta | 68.5% | val | Apr 5, 2025 | link ↗ |
Other model leaderboards
- gpt.buzz Composite Model Index — cross-benchmark ranking
- MMLU-Pro — Multitask language understanding across 14 disciplines — the harder, contamination-resistant successor to MMLU. 10-option multiple choice with stronger distractors.
- GPQA Diamond — PhD-level science questions in biology, physics, and chemistry, written by domain experts. The hardest subset of GPQA — humans with PhDs in the subject score ~65%.
- HumanEval — OpenAI's 164 Python programming problems with hidden unit tests — the classic code-generation pass@1 benchmark.
- Aider Polyglot — Multi-language code edit benchmark — solve real Exercism problems across Python, JavaScript, Rust, Go, Java, C++ via edit-and-test loops.
- AIME 2025 — American Invitational Mathematics Examination — competition math problems requiring multi-step reasoning. Reasoning-tier models score in the 90s; non-reasoning in the 50s.
- LiveCodeBench — Contamination-free code benchmark — only problems published AFTER a model's training cutoff. Refreshed monthly.