{
 "body": "\nPart of the [MMLU](mmlu.md) family.\n\n## What it measures\n\nThis id's meaning is not established. It appears on 40 frontier model cards here, pulled through\nthe llm-stats/intlpull enrichment with no notes, and none of the 40 carries any classic\nfour-option physics subject key (e.g. `mmlu_high_school_physics`). Two readings are possible.\n\nReading one: a mis-keyed MMLU-Pro Physics category score. MMLU-Pro has a ten-option Physics\ncategory (1,299 questions); classic 57-subject MMLU has no config named plain \"physics\".\nlm-evaluation-harness's own task name for that category, `mmlu_pro_physics`, is one segment from\nthis key, and on all 40 cards the value sits close in size to that model's `mmlu_pro` score.\n\nReading two: the \"physics\" STEM subcategory `categories.py` defines, pooling astronomy, college\nphysics, conceptual physics and high school physics (640 questions) -- the same rollup used here\nfor `mmlu_biology` and `mmlu_computer_science`. Against it: all 40 cards also separately report\n`mmlu_astronomy`, which that rollup would include.\n\n## Reading the numbers\n\nThe evidence leans toward reading one without confirming it -- the astronomy overlap is the\nstrongest fact against reading two, but nothing here traces the number to the upstream pull. Treat\nit as unverified: do not compare it to `mmlu_pro`, the classic physics subjects, or `mmlu_biology`\n/ `mmlu_computer_science` as if the grain matched, and do not fold it into a per-subject average.\nIt is pending a card re-key; check back before trusting this beyond a rough sense the model saw\nphysics questions.\n",
 "build": {
  "built_at": "2026-09-09T16:56:50+00:00",
  "commit": "0a599558854c0e238c03a0f0d725239cb28f9d11",
  "eligibility_as_of": "2026-09-09"
 },
 "disposition": {
  "canonical_id": "mmlu_physics",
  "reasons": [],
  "status": "unassessed",
  "verified_results": []
 },
 "models_covered": [
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Claude Opus 4",
   "model_id": "anthropic/claude-opus-4-20250514",
   "provider": "anthropic",
   "provider_display": "Anthropic",
   "score": 85.1,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Claude Opus 4.6",
   "model_id": "anthropic/claude-opus-4-6",
   "provider": "anthropic",
   "provider_display": "Anthropic",
   "score": 85.1,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, anthropic-system-card-mythos"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Gemini 2.5 Pro",
   "model_id": "google/gemini-2-5-pro",
   "provider": "google",
   "provider_display": "Google DeepMind",
   "score": 84.8,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "GPT-4.1",
   "model_id": "openai/gpt-4-1",
   "provider": "openai",
   "provider_display": "OpenAI",
   "score": 84.2,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, domain-evals preference-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "GPT-4o",
   "model_id": "openai/gpt-4o",
   "provider": "openai",
   "provider_display": "OpenAI",
   "score": 83.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "GPT-4o (2024-05-13)",
   "model_id": "openai/gpt-4o-2024-05-13",
   "provider": "openai",
   "provider_display": "OpenAI",
   "score": 83.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "GPT-4o (2024-08-06)",
   "model_id": "openai/gpt-4o-2024-08-06",
   "provider": "openai",
   "provider_display": "OpenAI",
   "score": 83.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "GPT-4o (2024-11-20)",
   "model_id": "openai/gpt-4o-2024-11-20",
   "provider": "openai",
   "provider_display": "OpenAI",
   "score": 83.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "GPT-4o mini",
   "model_id": "openai/gpt-4o-mini",
   "provider": "openai",
   "provider_display": "OpenAI",
   "score": 83.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, domain-evals, llm-stats, intlpull"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1",
   "model_id": "deepseek/deepseek-r1",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, preference-evals, open-llm-leaderboard-v2, domain-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 0528",
   "model_id": "deepseek/deepseek-r1-0528",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, preference-evals, open-llm-leaderboard-v2, domain-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 0528 NVFP4 v2",
   "model_id": "nvidia/deepseek-r1-0528-nvfp4-v2",
   "provider": "nvidia",
   "provider_display": "NVIDIA",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, preference-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 0528 Qwen3 8B",
   "model_id": "deepseek/deepseek-r1-0528-qwen3-8b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "bigcode-leaderboard, provider-reports, open-llm-leaderboard"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 Distill Llama 70B",
   "model_id": "deepseek/deepseek-r1-distill-llama-70b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 Distill Llama 8B",
   "model_id": "deepseek/deepseek-r1-distill-llama-8b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 Distill Qwen 1.5B",
   "model_id": "deepseek/deepseek-r1-distill-qwen-1-5b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "bigcode-leaderboard, provider-reports, open-llm-leaderboard, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 Distill Qwen 14B",
   "model_id": "deepseek/deepseek-r1-distill-qwen-14b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 Distill Qwen 32B",
   "model_id": "deepseek/deepseek-r1-distill-qwen-32b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek R1 Distill Qwen 7B",
   "model_id": "deepseek/deepseek-r1-distill-qwen-7b",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "DeepSeek Reasoner",
   "model_id": "deepseek/deepseek-reasoner",
   "provider": "deepseek",
   "provider_display": "DeepSeek",
   "score": 82.8,
   "source": "lmarena.ai, provider-reports, preference-evals, open-llm-leaderboard-v2, llm-stats"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Claude Sonnet 4",
   "model_id": "anthropic/claude-sonnet-4-20250514",
   "provider": "anthropic",
   "provider_display": "Anthropic",
   "score": 82.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Claude Sonnet 4.5",
   "model_id": "anthropic/claude-sonnet-4-5-20250929",
   "provider": "anthropic",
   "provider_display": "Anthropic",
   "score": 82.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Claude Sonnet 4.5 (latest)",
   "model_id": "anthropic/claude-sonnet-4-5",
   "provider": "anthropic",
   "provider_display": "Anthropic",
   "score": 82.5,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Qwen 3 235B Instruct",
   "model_id": "cerebras/qwen-3-235b-a22b-instruct-2507",
   "provider": "cerebras",
   "provider_display": "Cerebras",
   "score": 81.5,
   "source": "bigcode-leaderboard, provider-reports, preference-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Qwen3 235B-A22B",
   "model_id": "qwen/qwen3-235b-a22b",
   "provider": "qwen",
   "provider_display": "Alibaba / Qwen Team",
   "score": 81.5,
   "source": "lmarena.ai, provider-reports, preference-evals, llm-stats, domain-evals"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Gemma 4 31B",
   "model_id": "google/gemma-4-31b",
   "provider": "google",
   "provider_display": "Google DeepMind",
   "score": 78.8,
   "source": "lmarena.ai, provider-reports, multimodal-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "gemma 4 31B it",
   "model_id": "google/gemma-4-31b-it",
   "provider": "google",
   "provider_display": "Google DeepMind",
   "score": 78.8,
   "source": "lmarena.ai, provider-reports, multimodal-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "gemma 4 31B it GGUF",
   "model_id": "unsloth/gemma-4-31b-it-gguf",
   "provider": "unsloth",
   "provider_display": "Unsloth",
   "score": 78.8,
   "source": "lmarena.ai, provider-reports, multimodal-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Gemma 4 31B IT NVFP4",
   "model_id": "nvidia/gemma-4-31b-it-nvfp4",
   "provider": "nvidia",
   "provider_display": "NVIDIA",
   "score": 78.8,
   "source": "lmarena.ai, provider-reports, multimodal-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Mistral Large (latest)",
   "model_id": "mistral/mistral-large-latest",
   "provider": "mistral",
   "provider_display": "Mistral AI",
   "score": 78.2,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, open-llm-leaderboard-v2, llm-stats"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Mistral Large 2.1",
   "model_id": "mistral/mistral-large-2411",
   "provider": "mistral",
   "provider_display": "Mistral AI",
   "score": 78.2,
   "source": "lmarena.ai, provider-reports, safety-evals, preference-evals,, domain-evals open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Mistral Large 3",
   "model_id": "mistral/mistral-large-2512",
   "provider": "mistral",
   "provider_display": "Mistral AI",
   "score": 78.2,
   "source": "lmarena.ai, provider-reports, multimodal-evals, safety-evals, preference-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Gemma 4 26B",
   "model_id": "google/gemma-4-26b",
   "provider": "google",
   "provider_display": "Google DeepMind",
   "score": 77.5,
   "source": "lmarena.ai, provider-reports, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Llama 3.3 70B Instruct NVFP4",
   "model_id": "nvidia/llama-3-3-70b-instruct-nvfp4",
   "provider": "nvidia",
   "provider_display": "NVIDIA",
   "score": 76.8,
   "source": "lmarena.ai, provider-reports, safety-evals, preference-evals, open-llm-leaderboard-v2, llm-stats"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Llama-3.3-70B-Instruct",
   "model_id": "meta/llama-3-3-70b-instruct",
   "provider": "meta",
   "provider_display": "Meta",
   "score": 76.8,
   "source": "lmarena.ai, provider-reports, safety-evals, preference-evals, open-llm-leaderboard-v2, llm-stats"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Llama 3.1 70B",
   "model_id": "meta/llama-3-1-70b",
   "provider": "meta",
   "provider_display": "Meta",
   "score": 75.5,
   "source": "lmarena.ai, provider-reports, safety-evals, domain-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Llama 3.1 70B Instruct",
   "model_id": "meta/llama-3-1-70b-instruct",
   "provider": "meta",
   "provider_display": "Meta",
   "score": 75.5,
   "source": "lmarena.ai, provider-reports, safety-evals, domain-evals, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "phi 4",
   "model_id": "microsoft/phi-4",
   "provider": "microsoft",
   "provider_display": "Microsoft",
   "score": 74.2,
   "source": "lmarena.ai, provider-reports, preference-evals, open-llm-leaderboard-v2,, domain-evals llm-stats"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Phi 4 mini instruct",
   "model_id": "microsoft/phi-4-mini-instruct",
   "provider": "microsoft",
   "provider_display": "Microsoft",
   "score": 74.2,
   "source": "lmarena.ai, provider-reports, preference-evals, llm-stats, open-llm-leaderboard-v2"
  },
  {
   "as_of": "2026-04",
   "attribution": "unverified-legacy",
   "display_name": "Phi 4 multimodal instruct",
   "model_id": "microsoft/phi-4-multimodal-instruct",
   "provider": "microsoft",
   "provider_display": "Microsoft",
   "score": 74.2,
   "source": "lmarena.ai, provider-reports, preference-evals, open-llm-leaderboard-v2"
  }
 ],
 "page": {
  "aliases": [
   "mmlu_pro_physics"
  ],
  "category": "knowledge",
  "dataset": {
   "languages": [
    "en"
   ],
   "license": "MIT",
   "modalities": [
    "text"
   ],
   "public_test_set": true,
   "size_note": "Not established, because the underlying dataset depends on which reading is correct. If this is the MMLU-Pro Physics category (reading one), it would draw on 1,299 test questions from TIGER-Lab/MMLU-Pro. If this is the categories.py \"physics\" STEM subcategory (reading two), it would pool astronomy (152), college physics (102), conceptual physics (235) and high school physics (151) test questions from cais/mmlu -- 640 combined. Neither is confirmed for this id.\n"
  },
  "freshness": {
   "researched": "2026-09-08",
   "researched_by": "sonnet-5 agent, batch 1b, slice N"
  },
  "harness": {
   "other": "If this is the MMLU-Pro Physics category (reading one), the matching lm-evaluation-harness task is mmlu_pro_physics -- see the mmlu_pro page. If this is the categories.py rollup (reading two), the four pooled subjects run as separate tasks (mmlu_astronomy, mmlu_college_physics, mmlu_conceptual_physics, mmlu_high_school_physics) with no dedicated rollup task in any harness checked for this page. Neither is confirmed for this id.\n"
  },
  "id": "mmlu_physics",
  "lineage": {
   "family": "mmlu"
  },
  "measures": "Not established. This key appears on 40 frontier model cards in this repository with no notes on its source. It may be a mis-keyed MMLU-Pro Physics category score (MMLU-Pro has a ten-option Physics category; the original 57-subject MMLU has no subject or dataset config named plain \"physics\"), or it may be the \"physics\" STEM subcategory the original MMLU authors define in categories.py, which pools four classic subjects. See \"What it measures\" below for the evidence on each reading and why neither is confirmed.\n",
  "metric": {
   "baseline_note": "random_baseline is left empty because it depends on which reading is correct: 10% under MMLU-Pro's ten options (reading one), or 25% under the classic four-option MMLU format (reading two). Do not assume either without confirming the source first.\n",
   "direction": "higher_is_better",
   "max_score": 100,
   "name": "accuracy",
   "unit": "%"
  },
  "name": "MMLU: Physics (unresolved key)",
  "page_kind": "subset",
  "sources": [
   {
    "accessed": "2026-09-08",
    "title": "MMLU-Pro: A More Robust and Challenging Multi-Task Language Understanding Benchmark (Wang et al., arXiv:2406.01574)",
    "url": "https://arxiv.org/abs/2406.01574"
   },
   {
    "accessed": "2026-09-08",
    "title": "TIGER-Lab/MMLU-Pro dataset card, Hugging Face (14-category table, Physics = 1,299 questions)",
    "url": "https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro"
   },
   {
    "accessed": "2026-09-08",
    "title": "categories.py: MMLU subject-to-subcategory mapping, hendrycks/test repository",
    "url": "https://github.com/hendrycks/test/blob/master/categories.py"
   },
   {
    "accessed": "2026-09-08",
    "title": "cais/mmlu dataset card, Hugging Face",
    "url": "https://huggingface.co/datasets/cais/mmlu"
   },
   {
    "accessed": "2026-09-08",
    "title": "cais/mmlu datasets-server splits API (full 59-config list; no config named plain \"physics\")",
    "url": "https://datasets-server.huggingface.co/splits?dataset=cais/mmlu"
   },
   {
    "accessed": "2026-09-08",
    "title": "cais/mmlu datasets-server size API (row counts for astronomy, college_physics, conceptual_physics, high_school_physics)",
    "url": "https://datasets-server.huggingface.co/size?dataset=cais/mmlu"
   },
   {
    "accessed": "2026-09-08",
    "title": "lm-evaluation-harness mmlu_pro_physics task config",
    "url": "https://github.com/EleutherAI/lm-evaluation-harness/blob/main/lm_eval/tasks/mmlu_pro/mmlu_pro_physics.yaml"
   }
  ],
  "status": "unknown",
  "summary": "Disputed key on 40 frontier cards: possibly a mis-keyed MMLU-Pro Physics score, possibly a classic-MMLU STEM subcategory rollup. Neither reading is confirmed; pending a card re-key.",
  "tags": [
   "knowledge",
   "multiple-choice",
   "mmlu-subset",
   "unresolved"
  ],
  "task_format": "Not established -- it depends on which reading is correct. A ten-option, chain-of-thought format if this is an MMLU-Pro category score, or the classic four-option format if this is a categories.py subcategory rollup of four MMLU subjects.\n"
 }
}