FrontierSWE
Very long software engineering tasks at the edge of human ability. The score is the mean of five trials per task.
- Unit
- mean@5
- Results
- 21 results21 Tokscale models matched, 21 in the source
- Last update
- Oct 7, 2026Newest model release
- Retrieved
- Oct 21, 2026Source data through Oct 15, 2026
Conditions. One harness (proximus) for every model. Cost and duration are averages per task.
Attribution. FrontierSWE results, collected by Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks (CC BY 4.0)
Results
Every result in the source. The rank is the position in this list, by score; the source's own ranking can differ.
Tokscale usage counts accounts that used the model in public Tokscale submissions, Oct 14 – Nov 12, 2026. It is shown next to the score and does not explain it.
| Rank | Model | Score (mean@5) | Conditions | Cost and tokens | Date | Tokscale usage (30 days) | Source link |
|---|---|---|---|---|---|---|---|
| 1 | 65.5% | Max effort · proximus | $1,030 per task | Sep 3, 2026Model released | 562active users | Source for GPT-6 Astra (opens in a new tab) | |
| 2 | 62.3% | Max effort · proximus | $98.87 per task | Sep 22, 2026Model released | 709active users | Source for Claude Opus 5.5 (opens in a new tab) | |
| 3 | 61.9% | Max effort · proximus | $110 per task | Sep 28, 2026Model released | 555active users | Source for Claude Sonnet 5.5 (opens in a new tab) | |
| 4 | 56.3% | Max effort · proximusFrontierSWE V2 (mean@5 across 34 tasks, proximus harness, max reasoning effort). Site footnote: Claude Opus 5 used as fallback for tasks blocked by content filters. | $139 per task | Sep 1, 2026Model released | 361active users | Source for Claude Fable 5.1 (opens in a new tab) | |
| 5 | 55.0% | High effort · proximus | $129 per task | Sep 30, 2026Model released | 82active users | Source for Gemini 4 Argon (opens in a new tab) | |
| 6 | 52.0% | Max effort · proximus | $197 per task | Jul 24, 2026Model released | 163active users | Source for Claude Opus 5 (opens in a new tab) | |
| 7 | 47.0% | Max effort · proximus | $301 per task | Jun 9, 2026Model released | 41active users | Source for Claude Fable 5 (opens in a new tab) | |
| 8 | 43.8% | Max effort · proximus | $20.55 per task | Oct 7, 2026Model released | 477active users | Source for Claude Haiku 5.5 (opens in a new tab) | |
| 9 | GPT-5.6 SolListed as GPT-5.6 | 32.2% | Max effort · proximus | $180 per task | Jul 9, 2026Model released | 217active users | Source for GPT-5.6 Sol (opens in a new tab) |
| 10 | 30.2% | Max effort · proximus | $97.22 per task | Aug 14, 2026Model released | 138active users | Source for GLM-5.3 (opens in a new tab) | |
| 11 | 29.5% | XHigh effort · proximus | $319 per task | Sep 21, 2026Model released | 131active users | Source for Grok 4.7 (opens in a new tab) | |
| 12 | 25.9% | Max effort · proximus | $110 per task | Jul 16, 2026Model released | 56active users | Source for Kimi K3 (opens in a new tab) | |
| 13 | 25.8% | Max effort · proximus | $243 per task | Sep 2, 2026Model released | 21active users | Source for Muse Spark 1.3 (opens in a new tab) | |
| 14 | 25.3% | XHigh effort · proximus | $243 per task | Aug 12, 2026Model released | 41active users | Source for Grok 4.6 (opens in a new tab) | |
| 15 | 20.3% | High effort · proximus | $34.14 per task | Aug 13, 2026Model released | 4active users | Source for Gemini 3.7 Flash (opens in a new tab) | |
| 16 | 19.6% | High effort · proximus | $38.75 per task | Sep 2, 2026Model released | 66active users | Source for Gemini 3.8 Flash (opens in a new tab) | |
| 17 | 18.1% | Max effort · proximus | $11.03 per task | Aug 20, 2026Model released | 170active users | Source for GLM-5.3 Flash (opens in a new tab) | |
| 18 | Qwen3.8 Max 0902Listed as Qwen3.8-Max-0902 | 17.8% | XHigh effort · proximus | $51.27 per task | Sep 1, 2026Model released | 1active users | Source for Qwen3.8 Max 0902 (opens in a new tab) |
| 19 | Qwen3.8 MaxListed as Qwen3.8-Max | 15.8% | XHigh effort · proximus | $55.14 per task | Aug 2, 2026Model released | 16active users | Source for Qwen3.8 Max (opens in a new tab) |
| 20 | 12.0% | XHigh effort · proximus | $27.81 per task | Aug 5, 2026Model released | No Tokscale usage in the last 30 days | Source for Muse Spark 1.2 (opens in a new tab) | |
| 21 | 4.1% | XHigh effort · proximus | $9.15 per task | Jul 15, 2026Model released | No Tokscale usage in the last 30 days | Source for Inkling (opens in a new tab) |
Showing 21 of 21 results
Score by release date
Each dot is one model's best result on FrontierSWE, placed at the model's release date. The dashed line is the best score among models released up to that date.
- Used on Tokscale in the last 30 days
- No Tokscale usage in the last 30 days
- Best score so far
GPT-6 Astra: 65.5% · released Sep 3, 2026 · Max effort · proximus · 562 active users on Tokscale in the last 30 days
21 models, released Jun 9, 2026 to Oct 7, 2026. Highest: GPT-6 Astra, 65.5%, released Sep 3, 2026. Use the left and right arrow keys to read each model in release order.0%20%40%60%80%
Show data
| Model | Released | Score | Tokscale users, 30 days |
|---|---|---|---|
| Claude Haiku 5.5 | Oct 7, 2026 | 43.8%Max effort · proximus | 477 |
| Gemini 4 Argon | Sep 30, 2026 | 55.0%High effort · proximus | 82 |
| Claude Sonnet 5.5 | Sep 28, 2026 | 61.9%Max effort · proximus | 555 |
| Claude Opus 5.5 | Sep 22, 2026 | 62.3%Max effort · proximus | 709 |
| Grok 4.7 | Sep 21, 2026 | 29.5%XHigh effort · proximus | 131 |
| GPT-6 Astra | Sep 3, 2026 | 65.5%Max effort · proximus | 562 |
| Gemini 3.8 Flash | Sep 2, 2026 | 19.6%High effort · proximus | 66 |
| Muse Spark 1.3 | Sep 2, 2026 | 25.8%Max effort · proximus | 21 |
| Qwen3.8 Max 0902 | Sep 1, 2026 | 17.8%XHigh effort · proximus | 1 |
| Claude Fable 5.1 | Sep 1, 2026 | 56.3%Max effort · proximus | 361 |
| GLM-5.3 Flash | Aug 20, 2026 | 18.1%Max effort · proximus | 170 |
| GLM-5.3 | Aug 14, 2026 | 30.2%Max effort · proximus | 138 |
| Gemini 3.7 Flash | Aug 13, 2026 | 20.3%High effort · proximus | 4 |
| Grok 4.6 | Aug 12, 2026 | 25.3%XHigh effort · proximus | 41 |
| Muse Spark 1.2 | Aug 5, 2026 | 12.0%XHigh effort · proximus | No Tokscale usage in the last 30 days |
| Qwen3.8 Max | Aug 2, 2026 | 15.8%XHigh effort · proximus | 16 |
| Claude Opus 5 | Jul 24, 2026 | 52.0%Max effort · proximus | 163 |
| Kimi K3 | Jul 16, 2026 | 25.9%Max effort · proximus | 56 |
| Inkling | Jul 15, 2026 | 4.1%XHigh effort · proximus | No Tokscale usage in the last 30 days |
| GPT-5.6 Sol | Jul 9, 2026 | 32.2%Max effort · proximus | 217 |
| Claude Fable 5 | Jun 9, 2026 | 47.0%Max effort · proximus | 41 |
Method and matching
How results are run
- One harness (proximus) for every model. Cost and duration are averages per task.
- Results are Epoch AI's copy of the FrontierSWE leaderboard and can lag the live page.
Matching to Tokscale models
- 21 of 21 results are matched to 21 Tokscale models: 21 by exact id.
- No match is marked as reviewed by hand yet.
- Reasoning effort, thinking budget and agent are conditions of a result. They never change which model a result belongs to.
License and attribution
- CC BY 4.0 (opens in a new tab), retrieved Oct 21, 2026.
- FrontierSWE results, collected by Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks (CC BY 4.0)
Sources and attribution
Each score keeps its own source, unit and run conditions. Tokscale does not combine benchmarks into one score, and a missing score means the model was not evaluated, not that it scored zero.