How to read this chart
All 62 evaluation configurations are retained: 11 open-weight and 51 closed-weight results, representing 56 distinct model names. Scores and release dates come directly from Epoch’s underlying CSV, not estimated pixels.
A frontier result has no earlier-or-same-date result with an equal or higher mean score, except exact ties. Labels cover the overall frontier, the open-weight frontier, and the top three distinct open-weight score levels, including ties. Same-date, equal-score Astra configurations share a label.
Frontiers use reported means, not statistical significance. Error bars show ±1 standard error, not 95% confidence intervals. Results sharing coordinates overlap; inspect every configuration in the table.
What “open” means here
Open means publicly available model weights, using Epoch’s accessibility field with documented corrections for current availability. This is not a claim that training data, code, or licenses satisfy an open-source definition. Models without publicly available weights are classified as closed-weight.
GLM-5.3 is corrected to open weights. Its official model repository provides weights under the GLM-5.3 license. Epoch’s stale launch-time “API access” value is preserved separately in the table and downloads. The score remains 29.3%, tied with GLM-5.2; the earlier model retains the frontier position.
Kimi K3 is included. Epoch classifies its weights as non-commercial, but its note describes commercial-use thresholds requiring a separate agreement. It is better understood here as an open-weight model with license restrictions.
Dates describe model releases, not when evaluations were run. Accessibility reflects this snapshot and its documented corrections rather than a historical license reconstruction. Benchmark version: Tier 4 (v2), private set only.
Extracted data
62 results
| Model / evaluation setting | Release date | Accuracy | SE (pp) | Weight access | Original Epoch accessibility | Labels |
|---|---|---|---|---|---|---|
| GPT-6 Astra (high) | 2026-09-03 | 97.600% | ±2.400 | Closed weights | API access | Overall frontier |
| GPT-6 Astra (max) | 2026-09-03 | 97.600% | ±2.400 | Closed weights | API access | Overall frontier |
| GPT-6 Astra (xhigh) | 2026-09-03 | 97.600% | ±2.400 | Closed weights | API access | Overall frontier |
| GPT-6 Astra (medium) | 2026-09-03 | 97.561% | ±2.439 | Closed weights | API access | |
| Claude Fable 5 (max) | 2026-06-09 | 90.200% | ±4.600 | Closed weights | API access | Overall frontier |
| Claude Fable 5.1 (max) | 2026-09-01 | 87.805% | ±5.174 | Closed weights | API access | |
| GPT-6 Astra (low) | 2026-09-03 | 87.805% | ±5.174 | Closed weights | API access | |
| GPT-5.6 Sol (max) | 2026-07-09 | 82.927% | ±5.949 | Closed weights | API access | |
| GPT-6 Astra (none) | 2026-09-03 | 82.927% | ±5.949 | Closed weights | API access | |
| GPT-5.6 Sol (pro, max) | 2026-07-09 | 80.488% | ±6.266 | Closed weights | API access | |
| GPT-5.5 Pro (xhigh) | 2026-04-23 | 78.049% | ±6.545 | Closed weights | API access | Overall frontier |
| AI co-mathematician | 2026-05-08 | 75.600% | ±6.700 | Closed weights | Unreleased | |
| Claude Opus 5 (max) | 2026-07-24 | 73.171% | ±7.006 | Closed weights | API access | |
| GPT-5.5 (xhigh) | 2026-04-23 | 72.500% | ±7.150 | Closed weights | API access | |
| GPT-5.6 Terra (max) | 2026-07-09 | 70.732% | ±7.194 | Closed weights | API access | |
| GPT-5.6 Luna (max) | 2026-07-09 | 60.976% | ±7.713 | Closed weights | API access | |
| GPT-5.4 Pro (xhigh) | 2026-03-05 | 58.537% | ±7.790 | Closed weights | API access | Overall frontier |
| Claude Opus 4.8 (max) | 2026-05-28 | 56.098% | ±7.847 | Closed weights | API access | |
| GPT-5.4 (xhigh) | 2026-03-05 | 49.000% | ±8.000 | Closed weights | API access | |
| Qwen3.8 Max (xhigh) | 2026-08-02 | 46.341% | ±7.885 | Closed weights | API access | |
| GPT-5.2 Pro (xhigh) | 2025-12-11 | 46.000% | ±7.790 | Closed weights | API access | Overall frontier |
| Kimi K3 (Max) | 2026-07-16 | 39.024% | ±7.713 | Open weights | Open weights (non-commercial) | Open frontier; Top 3 open score levels (ties included) |
| Gemini 3.7 Flash (high) | 2026-08-13 | 36.585% | ±7.616 | Closed weights | API access | |
| Qwen3.7 Max | 2026-05-19 | 34.146% | ±7.498 | Closed weights | API access | |
| Qwen3.8 Max (0902) (xhigh) | 2026-09-01 | 34.146% | ±7.498 | Closed weights | API access | |
| Claude Opus 4.7 (max) | 2026-04-16 | 31.707% | ±7.358 | Closed weights | API access | |
| Grok 4.6 (xhigh) | 2026-08-12 | 31.707% | ±7.358 | Closed weights | API access | |
| GPT-5.2 (xhigh) | 2025-12-11 | 31.700% | ±7.300 | Closed weights | API access | |
| GLM-5.2 (max) | 2026-06-16 | 29.268% | ±7.194 | Open weights | Open weights (unrestricted) | Open frontier; Top 3 open score levels (ties included) |
| Claude Sonnet 5 (max) | 2026-06-30 | 29.268% | ±7.194 | Closed weights | API access | |
| GLM-5.3 (max) | 2026-08-14 | 29.268% | ±7.194 | Open weights (corrected) | API access | Top 3 open score levels (ties included) |
| Claude Opus 4.6 (max) | 2026-02-05 | 26.829% | ±7.006 | Closed weights | API access | |
| Gemini 3.1 Pro Preview | 2026-02-19 | 26.829% | ±7.006 | Closed weights | API access | |
| Gemini 3.5 Flash (high) | 2026-05-19 | 26.829% | ±7.006 | Closed weights | API access | |
| DeepSeek V4 Pro 0813 (max) | 2026-08-13 | 26.829% | ±7.006 | Open weights | Open weights (unrestricted) | Top 3 open score levels (ties included) |
| Kimi K2.6 | 2026-04-20 | 25.641% | ±7.083 | Open weights | Open weights (unrestricted) | Open frontier |
| Grok 4.5 (high) | 2026-07-08 | 24.390% | ±6.790 | Closed weights | API access | |
| DeepSeek V4 Flash 0731 (max) | 2026-07-31 | 24.390% | ±6.790 | Open weights | Open weights (unrestricted) | |
| GPT-5 (high) | 2025-08-07 | 21.951% | ±6.545 | Closed weights | API access | Overall frontier |
| Gemini 3.6 Flash (high) | 2026-07-21 | 21.951% | ±6.545 | Closed weights | API access | |
| Gemini 3.8 Flash (high) | 2026-09-02 | 21.951% | ±6.545 | Closed weights | API access | |
| GPT-5 Pro (high) | 2025-10-07 | 19.512% | ±6.266 | Closed weights | API access | |
| Gemini 3 Flash Preview | 2025-12-17 | 17.073% | ±5.949 | Closed weights | API access | |
| Grok 4.20 | 2026-02-17 | 17.073% | ±5.949 | Closed weights | API access | |
| Inkling Small (xhigh) | 2026-07-15 | 17.073% | ±5.949 | Open weights | Open weights (unrestricted) | |
| GLM-5.3-Flash (max) | 2026-08-20 | 17.073% | ±5.949 | Open weights | Open weights (unrestricted) | |
| Grok 4.3 Beta | 2026-04-17 | 14.634% | ±5.589 | Closed weights | Hosted access (no API) | |
| GPT-5 mini (high) | 2025-08-07 | 12.195% | ±5.174 | Closed weights | API access | |
| GPT-5.4 nano (high) | 2026-03-17 | 12.195% | ±5.174 | Closed weights | API access | |
| Kimi K2.7 Code | 2026-06-12 | 12.195% | ±5.174 | Open weights | Open weights (unrestricted) | |
| GPT-5.4 mini (xhigh) | 2026-03-17 | 9.756% | ±4.692 | Closed weights | API access | |
| o4-mini (high) | 2025-04-16 | 4.878% | ±3.406 | Closed weights | API access | Overall frontier |
| Claude Opus 4.5 (32k thinking) | 2025-11-24 | 4.878% | ±3.406 | Closed weights | API access | |
| Inkling (xhigh) | 2026-07-15 | 4.878% | ±3.406 | Open weights | Open weights (unrestricted) | |
| Claude Opus 4.1 (32k thinking) | 2025-08-05 | 2.439% | ±2.439 | Closed weights | API access | |
| GPT-5 nano (high) | 2025-08-07 | 2.439% | ±2.439 | Closed weights | API access | |
| Claude Sonnet 4.5 (32k thinking) | 2025-09-29 | 2.439% | ±2.439 | Closed weights | API access | |
| DeepSeek v4 (max) | 2026-04-24 | 2.439% | ±2.439 | Open weights | Open weights (unrestricted) | |
| GPT-5.5 Instant | 2026-05-05 | 2.439% | ±2.439 | Closed weights | API access | |
| o3-mini (high) | 2025-01-31 | 0.000% | ±0.000 | Closed weights | API access | Overall frontier |
| Gemini 2.5 Pro | 2025-06-17 | 0.000% | ±0.000 | Closed weights | API access | |
| Gemini 3.5 Flash-Lite (high) | 2026-07-21 | 0.000% | ±0.000 | Closed weights | API access |
Source: Epoch AI, CC BY as marked in the supplied SVG. Adapted visualization. Extracted 10 September 2026. Full raw precision is preserved in the downloadable CSV.