Skip to content
aiqbench

Capability frontier over time

Each Epoch benchmark, plotted on the release date of every model that's been scored on it. As newer models arrive the ceiling of what's been achieved keeps rising — a real trend, distinct from and often mistaken for individual models improving over time.

BenchmarkGPQA DiamondOTIS Mock AIMEFrontierMathSimpleQA VerifiedSWE-bench VerifiedARC-AGI-2Terminal-BenchHumanity's Last ExamSimpleBench

SWE-bench Verified — models on arrival

One point per model at its release date. The line traces the best-known score at each moment.

0%25%50%75%100%2024202520262027Model release dateScore (%)gpt-4o-2024-11-20 Released: May 2024 Score: 31.0% 95% CI: [28.9%, 33.1%] Date source: curatedclaude-3-7-sonnet-20250219 Released: Feb 2025 Score: 61.0% 95% CI: [58.7%, 63.2%] Date source: curatedgpt-4.1-2025-04-14 Released: Apr 2025 Score: 48.5% 95% CI: [46.3%, 50.8%] Date source: curatedo3-2025-04-16 Released: Apr 2025 Score: 62.3% 95% CI: [60.1%, 64.5%] Date source: curatedclaude-opus-4-20250514 Released: May 2025 Score: 70.7% 95% CI: [68.6%, 72.7%] Date source: curatedgemini-2.5-pro Released: Jun 2025 Score: 57.6% 95% CI: [55.3%, 59.8%] Date source: curatedclaude-opus-4-1-20250805 Released: Aug 2025 Score: 73.3% 95% CI: [71.3%, 75.4%] Date source: curatedgpt-5-2025-08-07 Released: Aug 2025 Score: 73.6% 95% CI: [71.5%, 75.6%] Date source: curatedgpt-5-mini-2025-08-07 Released: Aug 2025 Score: 64.7% 95% CI: [62.5%, 66.8%] Date source: curatedclaude-sonnet-4-5-20250929 Released: Sep 2025 Score: 71.3% 95% CI: [69.2%, 73.3%] Date source: curatedclaude-opus-4-5-20251101 Released: Nov 2025 Score: 76.7% 95% CI: [74.7%, 78.6%] Date source: name-suffixclaude-opus-4-5-20251101gpt-5.1-2025-11-13 Released: Nov 2025 Score: 68.0% 95% CI: [65.9%, 70.1%] Date source: name-suffixgpt-5.2-2025-12-11 Released: Dec 2025 Score: 73.8% 95% CI: [71.8%, 75.8%] Date source: name-suffixkimi-k2.5 Released: Jan 2026 Score: 73.8% 95% CI: [71.8%, 75.8%] Date source: huggingface-createdglm-5 Released: Feb 2026 Score: 72.1% 95% CI: [70.0%, 74.2%] Date source: huggingface-createdglm-5.1 Released: Apr 2026 Score: 74.2% 95% CI: [72.2%, 76.2%] Date source: huggingface-createdkimi-k2.6 Released: Apr 2026 Score: 76.7% 95% CI: [74.7%, 78.6%] Date source: huggingface-createdqwen3.7-max Released: May 2026 Score: 77.3% 95% CI: [75.4%, 79.2%] Date source: name-suffixqwen3.7-maxglm-5.2 Released: Jun 2026 Score: 78.7% 95% CI: [76.8%, 80.6%] Date source: huggingface-createdglm-5.2deepseek-v4-pro Released: Aug 2026 Score: 77.6% 95% CI: [75.7%, 79.5%] Date source: name-suffixgemini-3.5-flash Released: Sep 2026 Score: 79.3% 95% CI: [77.5%, 81.2%] Date source: lmarena-first-seengemini-3.5-flashclaude-opus-4-7 Released: Sep 2026 Score: 83.5% 95% CI: [81.8%, 85.2%] Date source: lmarena-first-seenclaude-opus-4-7gpt-5.4-2026-03-05 Released: Sep 2026 Score: 76.9% 95% CI: [74.9%, 78.8%] Date source: lmarena-first-seenclaude-sonnet-4-6 Released: Sep 2026 Score: 75.2% 95% CI: [73.2%, 77.2%] Date source: lmarena-first-seenclaude-opus-4-6 Released: Sep 2026 Score: 78.7% 95% CI: [76.9%, 80.6%] Date source: lmarena-first-seenqwen3.6-plus Released: Oct 2026 Score: 57.9% 95% CI: [55.6%, 60.1%] Date source: lmarena-first-seengpt-5.3-codex Released: Oct 2026 Score: 74.8% 95% CI: [72.8%, 76.8%] Date source: lmarena-first-seenqwen3.6-max-preview Released: Oct 2026 Score: 76.7% 95% CI: [74.7%, 78.6%] Date source: lmarena-first-seen
Model at its release dateRunning best score (frontier)Score confidence interval

Read this as a frontier chart, not a per-model trajectory. A fixed model's score on a fixed benchmark does not change over time — Claude 3.5 Sonnet's GPQA number today is the same one Anthropic reported at launch. The line rises because better models keep arriving, not because any specific model got better. Seeing one model's score change over time (as its evaluation is re-run with different scaffolds, for example) needs a different kind of data: daily snapshots of each board, which we will chart once several weeks of them have accumulated.

4 models have no known release date and are not plotted.

Benchmarks · frontier

Frontier milestones

Every time a ceiling moved: who, which benchmark, by how much. Tap a sparkline to pick a benchmark — the chart above follows it; tap an organisation to filter by lab. Each event is the diff of the running best the chart draws.

11 of 116 frontier events ·

  1. Sep 202679.3%

    gemini-3.5-flash Google DeepMind sets SWE-bench Verified

    +0.6 pp
  2. Sep 202683.5%

    claude-opus-4-7 Anthropic sets SWE-bench Verified

    +4.1 pp
  3. Jun 202678.7%

    glm-5.2 Z.ai (Zhipu AI) sets SWE-bench Verified

    +1.4 pp
  4. May 202677.3%

    qwen3.7-max Alibaba sets SWE-bench Verified

    +0.6 pp
  5. Nov 202576.7%

    claude-opus-4-5-20251101 Anthropic sets SWE-bench Verified

    +3.1 pp
  6. Aug 202573.6%

    gpt-5-2025-08-07 OpenAI sets SWE-bench Verified

    +0.2 pp
  7. Aug 202573.3%

    claude-opus-4-1-20250805 Anthropic sets SWE-bench Verified

    +2.7 pp
  8. May 202570.7%

    claude-opus-4-20250514 Anthropic sets SWE-bench Verified

    +8.3 pp
  9. Apr 202562.3%

    o3-2025-04-16 OpenAI sets SWE-bench Verified

    +1.4 pp
  10. Feb 202561.0%

    claude-3-7-sonnet-20250219 Anthropic sets SWE-bench Verified

    +30.0 pp
  11. May 202431.0%

    gpt-4o-2024-11-20 OpenAI sets SWE-bench Verified

    first score