ModelRefs / Benchmark Categories

Benchmark Categories

Browse AI benchmarks by category — reasoning, coding, agents, multimodal, safety, latency and cost — with the highest-authority benchmark in each.

Overview

Benchmarks are grouped by what they actually measure, because a score only means something next to another score from the same kind of test. Reasoning results and latency results are not on the same axis and should never be averaged into one number.

Each category below says what it covers and how many benchmarks are classified under it. Open one to see the individual benchmarks, what each measures, and the recorded scores with their sources.

All benchmark categories (11)

Categories group benchmarks that measure the same kind of capability, so results inside one are comparable with each other and results across two are not. 80 benchmarks are currently classified.

  • Reasoning (27)

    Graduate-level reasoning, math, and abstract problem solving. Measure depth-of-thought beyond pattern matching.

  • Coding (16)

    Code synthesis, repository-level engineering, and competitive programming. Predict how a model performs on real engineering work.

  • Vision (6)

    Image understanding, OCR, document parsing, and visual QA. Evaluate visual perception and grounded interpretation.

  • Open Source (6)

    Open-weight model rankings, license comparison, and self-host viability. Find the best open alternatives to closed frontier models.

  • Multimodal (6)

    Cross-modal reasoning across text, image, audio, and video. Benchmark models that operate across modalities at once.

  • Retrieval (5)

    Embedding and retrieval benchmarks for semantic search and multilingual RAG. Evaluate retrieval-oriented representation quality before application-specific RAG testing.

  • Agents (5)

    Tool-use, function-calling, web/computer-use, and multi-step task execution. Measure end-to-end agentic task completion under real environments.

  • Latency (4)

    First-token latency, tokens/sec, streaming throughput. Quantify real-world responsiveness for interactive UX.

  • Safety (3)

    Red-team, jailbreak, toxicity, and refusal-quality evaluations. Quantify resistance to misuse and harmful-output rate.

  • Pricing (1)

    Input + output token pricing across providers. Compare $/Mtok across the frontier.

  • Cost Efficiency (1)

    Quality-per-dollar across coding, reasoning, and latency. Find the best value model for a given workload.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Benchmark Categories.