EN
EnglishEspañol中文PortuguêsFrançaisالعربية日本語한국어

Updated Oct 10, 2026

Model comparison

The Latent benchmark database

Published benchmark results for 31 frontier models, kept in each benchmark's native units and grouped into 7 capability domains.

These results are scattered across lab announcements, leaderboards, and PDFs, and no two of them are reported the same way. This page gathers the benchmarks worth comparing into one table, exactly as their publishers reported them.

205benchmarks tracked
92with enough coverage to compare
113composites + excluded evidence
31frontier models tracked

15 benchmarks on this page

Professional & real-world work

Economically valuable knowledge work: legal, finance, tax, medical, enterprise documents, and occupational workflows.

BenchmarkOpus 5.5AnthropicGPT-6 AstraOpenAIGemini 4 ArgonGoogleFable 5.1AnthropicSonnet 5.5AnthropicOpus 5AnthropicMiMo-V2.6-ProXiaomiFable 5AnthropicGPT-6.1 SolOpenAIGPT-6 SolOpenAIMuse Spark 1.3MetaDeepSeek V4.1-FlashDeepSeekGPT-5.6 SolOpenAIGrok 4.7xAIGrok 4.6xAIKimi K3Moonshot AIGLM-5.3Z.aiGemini 3.8 FlashGoogleGemini 3.7 FlashGoogleClaude Haiku 5.5AnthropicGPT-6 LunaOpenAIGPT-5.6 TerraOpenAIGLM 5.3 FlashZ.aiQwen3.8-MaxAlibabaMuse Spark 1.2MetaDeepSeek V4 ProDeepSeekMistral Large 4Mistral AIGPT-5.6 LunaOpenAISonnet 5AnthropicQwen3.8-27BAlibabaDeepSeek V4 FlashDeepSeekEvidence
GDPval-AA v2 (Elo)Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-1,580.2-1,763.64-1,738.08-1,741--1,719.65-1,627.48-1,665.731,587.51,679.471,464.961,435.34--1,482.49-1,633.531,527.081,496.68-1,492.731,504.871,466.211,580.66—
Harvey LAB (Vals)Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-5.42%19.6%6.67%-6.67%-11.25%----2.5%-15.8%10.83%8.33%10%8.75%--0.83%-10.42%25.42%7.5%-1.25%5%11.25%8.33%SourceCitation
HealthBench Professional (length-adjusted)Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-63.4%---56.4%-60.9%----60.5%----52.1%---57.7%-----55.7%---—
Internal Data Science Tasks - OpenAICurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-40.9%-----34.7%----30.5%------------------—
Internal Design Tasks - OpenAICurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-50%-----35.8%----47.4%------------------—
JobBenchCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-----65.7%62%57.4%--61.2%-45.4%--54.3%-------53.4%61.6%----33.4%-SourceCitation
Legal Research Bench - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-39.42%-55.29%-55.29%-49.52%----48.08%-48.08%44.23%49.04%38.94%34.62%--41.35%-47.6%43.75%40.87%-36.54%41.83%36.06%30.29%—
LegalBench - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.---88.51%-86.97%-88.56%----86.97%-86.31%86.02%84.84%86.99%87.26%--85.11%-83.61%85.26%82.36%-84.03%83.92%82.43%77.71%SourceCitation
MedCode - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-48.49%-53.51%-63.57%-56.07%----43.97%-44.71%48.88%42.86%48.13%53.39%--43.41%-40.67%49.35%42.47%-42.39%47.54%28.7%41.41%SourceCitation
MedScribe - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-87.91%-91.29%-90.98%-88.52%----85.23%-86.53%87.96%88.81%84.5%83.94%--82.87%-84.95%90.06%80.17%-84.39%76.05%83.85%80.36%SourceCitation
MortgageTax - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.---70.79%-72.06%-68.92%----67.29%-64.19%66.34%-65.34%66.65%--67.33%-63.99%65.42%--67.29%70.03%64.94%-—
OfficeQA Pro †Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.67.7%--69%65.6%66.9%-69.9%----63.2%-63.2%63.3%------------59.4%--—
Public Benefits Bench - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.---74.9%-76.93%-70.43%----66.51%-66.85%68.27%68.54%65.29%---62.38%-67.12%68.47%62.92%-61.16%66.03%--SourceCitation
Tax Agent Bench v1 - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-63.34%-77.64%------71.93%-67.95%-70.79%68.67%73.09%66.77%57.66%--65.2%---58.66%-60.81%62.27%--—
TaxEval v2 - ValsCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.---75.96%-75.14%-76.94%----74.78%-71.1%75.72%72.36%74.45%74.73%--76.17%-75.55%80.38%73.06%-76.17%75.63%70.85%70.69%SourceCitation

5 benchmarks on this page

Multimodal & vision

Chart, image, video, and handwriting understanding.

BenchmarkOpus 5.5AnthropicGPT-6 AstraOpenAIGemini 4 ArgonGoogleFable 5.1AnthropicSonnet 5.5AnthropicOpus 5AnthropicMiMo-V2.6-ProXiaomiFable 5AnthropicGPT-6.1 SolOpenAIGPT-6 SolOpenAIMuse Spark 1.3MetaDeepSeek V4.1-FlashDeepSeekGPT-5.6 SolOpenAIGrok 4.7xAIGrok 4.6xAIKimi K3Moonshot AIGLM-5.3Z.aiGemini 3.8 FlashGoogleGemini 3.7 FlashGoogleClaude Haiku 5.5AnthropicGPT-6 LunaOpenAIGPT-5.6 TerraOpenAIGLM 5.3 FlashZ.aiQwen3.8-MaxAlibabaMuse Spark 1.2MetaDeepSeek V4 ProDeepSeekMistral Large 4Mistral AIGPT-5.6 LunaOpenAISonnet 5AnthropicQwen3.8-27BAlibabaDeepSeek V4 FlashDeepSeekEvidence
BabyVision (with CI)Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-------90.5%----88.9%--85.7%-------91.3%-----85.6%-—
CharXiv (with CI / RQ)Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-----89.3%-93.5%----89.1%--91.3%-86.2%88.7%--88%89.4%93.5%----88.3%90.2%-—
MathVision (with CI)Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-------98.6%----97.8%--97.8%-------97.7%-----94.6%-—
MMMU-ProCurrent values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.87.69%86.88%-90.64%-84.74%-81.2%85.95%83.29%82.02%-83.41%--80.52%-85.61%85.49%-79.71%80.69%-82.31%--76.42%78.55%77.28%76.3%-SourceCitation
OmniDocBench 1.5Current values follow the best-published-reasoning-mode policy. Individual source, configuration and admission decisions are disclosed on /methodology#model-evidence. Collected values without a matching review remain visible but do not contribute.-------89.5%----85.8%--91.1%-------92.1%-----91.1%-—

Reading this table

Native units, nothing adjusted

Units

Cells show each benchmark's native scale: accuracy percentages, Elo ratings, or points. Nothing is rescaled or converted, so compare along a row. Two different rows are two different scales.

Best in row

The strongest published result in each row is highlighted. A blank cell means that model has no published result on that benchmark, which is not the same as a zero.

What gets listed

A benchmark earns a row when its result is published against a named model version with a stated metric, and when enough models have run it for the comparison to mean something. Numbers we cannot trace back to a source are left out.

Provenance

Results are current as of 2026-10-10. Vendor-reported head-to-head rows are named as such (e.g. “Anthropic H2H”) and carry their caveats in the row notes.

Evidence downloads: manifest · dataset JSON · results CSV · citations CSV