AI model benchmarks

Leaders by category

The top spot of every benchmark with data. Compare them only within the same benchmark.

Frontier models

At Promptea, a model is “frontier” when it ranks in the top 5 of at least one general-capability benchmark in the current snapshot: Artificial Analysis' Intelligence, Coding or Agentic index, or OpenRouter's GPQA Diamond and τ-bench runs.

We use these because they measure broad capability — reasoning, coding and agentic work. Specialized rankings (search configurations, design arenas) are shown, but a narrow win there is not enough to call a model “frontier”. The list changes when the data does.

• Claude Fable 5.1 (Max, Default Fallback)

  Anthropic

  • #3 in Intelligence Index
  • #1 in Coding Index
  • #1 in Agentic Index
  • #3 in τ-bench (verified, airline)

  Prompt guideFind the best AI

• Gemini 4 Argon (High)

  Google

  • #5 in Intelligence Index
  • #5 in τ-bench (verified, airline)
  • #1 in GPQA Diamond

  Prompt guideFind the best AI

• Claude Opus 5 (Max)

  Anthropic

  • #2 in Coding Index
  • #2 in Agentic Index
  • #4 in τ-bench (verified, airline)

  Prompt guideFind the best AI

• GPT-6 Astra (Max)

  OpenAI

  • #4 in Intelligence Index
  • #4 in Coding Index

  Prompt guideFind the best AI

• Anthropic: Claude Fable 5

  Anthropic

  • #1 in τ-bench (verified, airline)

  Prompt guideFind the best AI

• Claude Opus 5.5 (Max, Default Fallback)

  Anthropic

  • #1 in Intelligence Index

  Prompt guideFind the best AI

• Claude Sonnet 5.5 (Max, Default Fallback)

  Anthropic

  • #2 in Intelligence Index

  Prompt guideFind the best AI

• Google: Gemini 3.8 Flash

  Google

  • #2 in GPQA Diamond

  Prompt guideFind the best AI

• inclusionAI: Ling 3.1 Flash

  Inclusionai

  • #2 in τ-bench (verified, airline)

• GPT-5.6 Sol (Max)

  OpenAI

  • #3 in Coding Index

  Prompt guideFind the best AI

• OpenAI: GPT-6 Astra Pro

  OpenAI

  • #3 in GPQA Diamond

  Prompt guideFind the best AI

• Qwen3.8 Max

  Qwen

  • #3 in Intelligence Index

• Qwen3.8 Max (0902)

  Qwen

  • #3 in Agentic Index

• Fugu Ultra

  Sakana

  • #4 in GPQA Diamond

• Muse Spark 1.3 (Max)

  Meta

  • #4 in Agentic Index

• GLM-5.3 (Max)

  Z.ai

  • #5 in Agentic Index

• Google: Gemini 3.1 Pro Preview

  Google

  • #5 in GPQA Diamond

  Prompt guideFind the best AI

• Grok 4.6 (High)

  xAI

  • #5 in Coding Index

  Prompt guideFind the best AI

Explore the rankings

What each benchmark measures

Intelligence Index · General intelligence

Measures: General capability: aggregates reasoning, knowledge, math and coding evaluations into a single index.

Limitations: It is a composite index: it combines several third-party evaluations with weights chosen by Artificial Analysis, and it changes between versions. Use it to compare models within the same index, not to predict results on your specific task.

Source: Artificial Analysis · Composite index (higher is better)

Coding Index · Coding

Measures: Coding: aggregates code generation and problem-solving evaluations.

Limitations: It is a composite index: it combines several third-party evaluations with weights chosen by Artificial Analysis, and it changes between versions. Use it to compare models within the same index, not to predict results on your specific task.

Source: Artificial Analysis · Composite index (higher is better)

Agentic Index · Agents and tools

Measures: Agentic work: aggregates tool-use and multi-step task evaluations.

Limitations: It is a composite index: it combines several third-party evaluations with weights chosen by Artificial Analysis, and it changes between versions. Use it to compare models within the same index, not to predict results on your specific task.

Source: Artificial Analysis · Composite index (higher is better)

τ-bench (verified, airline) · Agents and tools

Measures: An agent serves a simulated airline customer using tools while following policies. Run by OpenRouter.

Limitations: Covers a single domain (airline) with a simulated user; it does not measure other kinds of agents or real integrations.

Source: OpenRouter · Accuracy: % of tasks solved

GPQA Diamond · Scientific reasoning

Measures: Graduate-level multiple-choice questions in biology, physics and chemistry that are hard to answer by searching the web. Run by OpenRouter.

Limitations: Multiple choice and single-domain (science); top models already score very high, so small gaps can fall within run-to-run variation.

Source: OpenRouter · Accuracy: % correct answers

BrowseComp (search) · Search

Measures: OpenRouter search benchmark: the model answers using a web-search tool.

Limitations: OpenRouter publishes each model's best-scoring configuration (search engine, surface, effort), so the result also depends on that configuration, not only on the model.

Source: OpenRouter · Accuracy: % correct answers

DSQA (search) · Search

Measures: OpenRouter search benchmark: the model answers using a web-search tool.

Limitations: OpenRouter publishes each model's best-scoring configuration (search engine, surface, effort), so the result also depends on that configuration, not only on the model.

Source: OpenRouter · Accuracy: % correct answers

Humanity's Last Exam (search) · Search

Measures: OpenRouter search benchmark: the model answers using a web-search tool.

Limitations: OpenRouter publishes each model's best-scoring configuration (search engine, surface, effort), so the result also depends on that configuration, not only on the model.

Source: OpenRouter · Accuracy: % correct answers

WideSearch (search) · Search

Measures: OpenRouter search benchmark: the model answers using a web-search tool.

Limitations: OpenRouter publishes each model's best-scoring configuration (search engine, surface, effort), so the result also depends on that configuration, not only on the model.

Source: OpenRouter · F1 by item (higher is better)

Design Arena · 3d · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · asciiart · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · audiorealism · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · codecategories · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · dataviz · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · gamedev · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · graphicdesign · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · image · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · imageediting · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · logo · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · svg · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · text to speech · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · uicomponent · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)

Design Arena · website · Design and UI

Measures: Head-to-head comparisons voted by people: which visual result they prefer between two models for the same brief.

Limitations: Measures preference, not correctness. Elo is only comparable within the same arena and category.

Source: Design Arena · Elo (higher is better; relative to this arena)