Other
We track 41 benchmarks in other, or 50 if you count each version. They run from everyday level to research frontier level.
Anthropic held the top other score 46% of the contested time
That's about 21 of the 46 months since January 2023, averaged over the benchmarks below. The easiest benchmarks are at the top. Versions of one test share a row, with a dark tick at each new version. We stop the clock, and fade the rest of the row, when the best score passes 85%, or at the newest score if 45 days go by without another.
Anthropic OpenAI Google DeepMind
University Undergraduates and entry-level professionals
Expert Specialists with years of training, often over hours of work
Share of contested time at number one
Anthropic 46% OpenAI 43% 
Google DeepMind 11%
Google DeepMind 11% Scores come from each benchmark’s own leaderboard, from Epoch AI’s runs and from labs’ reports on their own models. We date Epoch runs by the model’s release and lab reports by when they came out. Labs pick their own settings, so a lab’s claim can beat an independent run of the same model. Turn off lab claims to see independent runs only.
Anthropic OpenAI Google DeepMind
University Undergraduates and entry-level professionals
Expert Specialists with years of training, often over hours of work
Share of contested time at number one, independent runs only
Anthropic 48% Google DeepMind 40% OpenAI 12%
Google DeepMind 40% OpenAI 12% Only each benchmark’s own leaderboard and Epoch AI’s runs, dated by the model’s release. We leave out benchmarks that only labs report.
From everyday to research frontier level
Grouped by who could solve the tasks, from things most adults can do to open research problems. The bar under each name is its best score.
Everyday
Most adults, without training
University
Undergraduates and entry-level professionals
Expert
Specialists with years of training, often over hours of work
Research frontier
Researchers, on open problems or tasks that take days
Another 13 have no level. Levels don’t apply to arenas, safety evaluations or composite indices.
By kind
Composite indices
| Benchmark | Level | Scored | Best score | Status |
|---|---|---|---|---|
| OpenVLM Leaderboard OpenCompass runs vision-language models through VLMEvalKit on a fixed set of image benchmarks, such as MMBench, MMStar and MathVista. The headline is the average score across that set. | University | Checked automatically | 80.1 | Not in cards |
| Conceptual Reasoning Index An index that ranks models on reasoning about hard-to-verify topics such as AI alignment and decision theory. It weights LMCA 60 percent and ACCoRD and DTBench 20 percent each. | Expert | Checked automatically | 77.9 | Not in cards |
| HELM Lite A lightweight HELM board that runs models with in-context learning on ten scenarios. They include NarrativeQA, NaturalQuestions, OpenbookQA, MMLU, GSM8K, MATH, LegalBench, MedQA and WMT 2014. | University | Checked automatically | 93.8 | Not in cards |
| CAIS Text Capabilities Index Averages model scores on the text tests that the Center for AI Safety runs for its AI Dashboard. On 2026-11-01 the index listed Humanity's Last Exam and TextQuests. | Research frontier | Checked automatically | 66.5 | Not in cards |
| CAIS Risk Index Averages seven safety measures that correlate weakly with general capability. They cover agent prompt injection, bioweapons help, political bias, overconfidence, lying under pressure and harmful or deceptive play in text games. | Research frontier | Checked automatically | 26.2 | Not in cards |
| HELM Capabilities Stanford CRFM runs language models on a curated set of scenarios, one per key capability, and averages the scores. Prompt-level results are public. | University | Checked automatically | 81.9 | Not in cards |
| CAIS Vision Capabilities Index Averages model scores on five vision tests run by the Center for AI Safety: ERQA, MindCube, SpatialViz-Bench, IntPhys 2 and EnigmaEval. | Research frontier | Checked automatically | 83.3 | Not in cards |
| Vals Multimodal Index Models are scored on a GDP-weighted mix of seven Vals finance, coding and education benchmarks, including MortgageTax and SAGE, which grades handwritten math work. | Expert | Graded by a model | 74.2% | Not in cards |
| Vals Science Index Agents are scored on the equal average of BioMysteryBench, MysteryMechanism and Terminal-Bench-Science. | Expert | Checked automatically | 65.1% | Not in cards |
| SuperCLUE The CLUE team tests Chinese and foreign language models on periodic sets of Chinese questions covering language, professional knowledge, reasoning, agents and safety. Open answers are graded by a strong LLM judge, and results feed an overall ranking. | University | Graded by a model |  72.6 | Not in cards |
| HELM Classic Models are evaluated with standardized prompts on the scenarios from the original HELM paper. Each scenario reports accuracy, calibration, robustness, fairness, bias, toxicity and efficiency. | University | Checked automatically |  94.3 | Not in cards |
| Hermes Index Nous Research averages pass@1 and cost per task over Hermes Bench, Terminal-Bench 4.0, Terminal-Bench Science and SkillsBench, all run in Hermes Agent. | Expert | Checked automatically | 63.3 | Not in cards |
| Artificial Analysis Intelligence Index v4.1 Version 4.1 of Artificial Analysis's index, which moved to GDPval-AA v2, Terminal-Bench 2.1 and τ³-Banking and put the most weight on agents. | Checked automatically | 61.0 | In use | |
| Artificial Analysis Intelligence Index v4.0 Version 4.0 of Artificial Analysis's index, ten evaluations in four equal categories that brought in real-world knowledge work and hallucination tests. | Checked automatically | 60.2 | Fading | |
| Artificial Analysis Intelligence Index v4.3 Version 4.3 of Artificial Analysis's index, eleven weighted parts led by knowledge work, office tasks and Terminal-Bench 4.0. | Checked automatically | 56.0 | In use | |
| Anthropic ECI (AECI) Index that combines a basket of capability benchmarks into one capability score to track trends across Anthropic models. | Research frontier | Checked automatically | 169.4 | System cards only |
| Artificial Analysis Intelligence Index v2 Artificial Analysis's index of seven public reasoning, knowledge, math and coding benchmarks, each run by Artificial Analysis. | Checked automatically | 67.7 | Dropped | |
| Artificial Analysis Intelligence Index v3 Version 3 of Artificial Analysis's index, ten evaluations that added agentic terminal and tool-use tasks to the knowledge, math and coding tests. | Checked automatically |  61.0 | Dropped | |
| Decision Index Decision models read a state and a typed question and return a decision. The index weights private same-skill sets and new-domain sets, equated on 21 base models. | Everyday | Checked automatically | 67.6 | Not in cards |
| AGI Score (A Definition of AGI) Experts test models on ten cognitive domains from Cattell-Horn-Carroll theory, such as reasoning, memory and perception, with adapted human psychometric tests. Each domain counts 10% toward a score out of 100%. | Expert | Graded by people | 57% | Not in cards |
| Artificial Analysis Intelligence Index v5 Version 5 of Artificial Analysis's index: nine evaluations in four equal categories, most with private test sets, run three times each in its own agent harness. | Checked automatically | 59.0 | In use | |
| Personal Superintelligence Index (Meta) Meta's composite of seven personal and knowledge-work evals: Agents' Last Exam, AutomationBench-AA, FocusBench, JobBench, GDP.pdf, VBT and DRACO. | Expert | Checked automatically | In use | |
| Artificial Analysis Decision Index Decision models answer structured questions with a probability for each option. Four evaluations weighted equally cover classification, tool calling, knowledge with hallucination, and reasoning. | Everyday | Checked automatically | Not in cards | |
| Artificial Analysis Intelligence Index v1 Artificial Analysis's first composite score, called the Quality Index in 2024, which normalized and combined a few public benchmarks into one number. | Checked automatically | Not in cards | ||
| Artificial Analysis Intelligence Index v2.2 Version 2.2 of Artificial Analysis's index, which swapped the older math tests for AIME 2025 and added instruction following and long-context reasoning. | Checked automatically | Not in cards | ||
| Artificial Analysis Intelligence Index v4.2 Version 4.2 of Artificial Analysis's index, which added multi-hour office work and document tasks and dropped GPQA Diamond. | Checked automatically | Not in cards | ||
| Artificial Analysis Progress Index Artificial Analysis's index that tracks how model capability changes over time on one continuous scale, launched alongside Intelligence Index v5. | Checked automatically | Not in cards | ||
| CompassRank OpenCompass ranks open and API language models on its own evaluation sets and publishes monthly rankings. Model makers submit a repository link or API endpoint to be tested. | Expert | Checked automatically | Not in cards | |
| Epoch Capabilities Index An index that fits model capability and benchmark difficulty from scores on over 50 benchmarks in Epoch AI's hub, using item response theory. Scores read like Elo ratings. | Research frontier | Checked automatically | Not in cards | |
| Fireworks Specialized Intelligence Index An index that gathers 66 practitioner-built benchmarks in 14 domains, such as legal, finance, healthcare and software. It combines score, cost and duration per benchmark into one number. | Expert | Graded by a model | Not in cards | |
| Open LLM Leaderboard v2 Hugging Face ran open models on IFEval, BBH, MATH Level 5, GPQA, MuSR and MMLU-Pro and averaged normalized scores. | University | Checked automatically | Not in cards | |
| Vals Router Index Model routers run all eight Vals Index v2.2 benchmarks through the same agents as single models and are scored with the Index weights. | Expert | Graded by a model | Not in cards |
Other
| Benchmark | Level | Scored | Best score | Status |
|---|---|---|---|---|
| ACCoRD Models state probabilities and preferences on conceptual questions, and the benchmark checks 567 expert-approved logical consistency constraints such as conjunction rules and order invariance. | Expert | Checked automatically | 86.5 | Not in cards |
| LMCA Models rate arguments against position texts on philosophy, decision theory, and AI risk. Scored by agreement with expert ratings, with a ceiling near 85. | Expert | Checked automatically | 68.3% | Not in cards |
| LiveBench Models answer monthly refreshed questions across math, coding, reasoning, language, instruction following, and data analysis. Scored automatically against objective ground truth. | University | Checked automatically | 85.7% | Dropped |
| SAGE (Vals) Models grade handwritten undergraduate work in PDEs, linear algebra and computer science against rubrics. Scores compare their deductions with expert graders. | University | Graded by people | 56.1% | Not in cards |
| Political Science LLM Benchmark 33 political-science text-coding tasks, 100 texts each (~3,300 items). Models scored by mean F1 against human-coded labels. | University | Checked automatically | 71.4% | Not in cards |
| SAGE v1.1 (Vals) Models grade handwritten undergraduate math and computer science work against rubrics. Scored by class-balanced accuracy of each rubric check against expert grading. | University | Graded by people | 91.3% | Not in cards |
| PhiBench Models answer original prompts that Microsoft's Phi team wrote to test a wide range of skills and reasoning. | University | Graded by a model |  74.2% | Dropped |
| Pile-test Models predict text from the test split of the Pile, an 800GB corpus of 22 sources. Scored by bits per byte. | Checked automatically |  0.55 | Dropped | |
| Qwen in-house English automatic evaluation Models answer Qwen in-house English datasets on knowledge, comprehension, coding, math and related skills. Skills and weights change between reports. | Everyday | Graded by a model | 72.9% | System cards only |
| Recognition of model-written user turns (Anthropic) Models write synthetic user turns for real Claude Code transcripts, then judge which version is real. Score is accuracy above chance. | Expert | Checked automatically | 0.35 | System cards only |
| JevBench Decision models, such as TypeSafe AI's Jev and compatible systems, answer typed decision questions. The board scores accuracy, probability calibration, latency and cost. | Everyday | Checked automatically | Not in cards | |
| MA-Bench Models do military mission analysis under the joint planning process on long-context scenarios, graded against expert rubrics. | Expert | Graded by a model | Not in cards | |
| MTEB Massive Text Embedding Benchmark: retrieval, clustering, classification across languages and modalities. | Checked automatically | Not in cards | ||
| MTEB Arena Users compare retrieval, clustering and similarity results from two anonymous embedding models and vote for the better one. | Everyday | Ranked by votes | Not in cards | |
| RTEB Embedding models retrieve documents on open and private datasets across 20 languages and domains such as law, healthcare, code and finance. Scored by NDCG@10. | University | Checked automatically | Not in cards | |
| StudentBench AI tutors teach real students GRE material and are scored by the students measured learning gains. A second study has expert tutors compare AI lesson plans and practice problems. | University | Settled by events | Not in cards | |
| TutorBench Models act as tutors on three tasks: adaptive explanations for a confused student, feedback on student work, and hints. Graded against sample-specific rubrics. | University | Graded by a model | Not in cards | |
| Xbench Measures firsthand X posts about frontier models and coding harnesses. An LLM labels each post's sentiment, aggregated per model per week. | Ranked by votes | Not in cards |