AI Model Capability Comparison
Agentic coding
Only models released within the past year are shown. Only benchmarks covering at least 3 of these models are shown. Unknown release dates are excluded; historical data is retained.
— means no published score for this candidate in this source, not zero.
| No. | Model | 4/11 recent-year models have scores | 4/11 recent-year models have scores | 4/11 recent-year models have scores | 4/11 recent-year models have scores |
|---|---|---|---|---|---|
| 1 |  Claude Opus 5.5 | 74.2Claude Opus 5.5 | 62.3Claude Opus 5.5 | 66.4Claude Opus 5.5 | 90.3Claude Opus 5.5 |
| 2 |  GPT-6 Astra | 74.1GPT-6 Astra | 65.5GPT-6 Astra | 58.2GPT-6 Astra | 89.6GPT-6 Astra |
| 3 | GPT-6.1 Sol | — | — | — | — |
| 4 | Claude Fable 5.1 | 67.4Claude Fable 5.1 | 56.3Claude Fable 5.1 | 57.9Claude Fable 5.1 | 90.3Claude Fable 5.1 |
| 5 |  Gemini 4 Argon | 77.9Gemini 4 Argon | 55Gemini 4 Argon | 57.4Gemini 4 Argon | 91.9Gemini 4 Argon |
| 6 | Claude Sonnet 5.5 | — | — | — | — |
| 7 | Claude Opus 5 | — | — | — | — |
| 8 | GPT-6 Sol | — | — | — | — |
| 9 | GPT-6 Luna | — | — | — | — |
| 10 | Claude Haiku 5.5 | — | — | — | — |
| 11 | Claude Mythos Preview | — | — | — | — |
Google DeepMind (model card)
Source: Google DeepMind (model card) ↗
Google model card: publisher results and republished third-party scores, as of October 2026. See methodology for evaluation settings; these are not all independent evaluations. Evaluation methodology ↗
DeepSWE v1.1 · Source date unknown
FrontierSWE v2 · Source date unknown
Terminal-bench 4.0 · Source date unknown
Vibe Code Bench · Source date unknown
PandaNpc Updated Oct 21, 2026, 06:01 UTC
Source: Epoch AI, CC BY 4.0 · Vals.ai · Artificial Analysis (artificialanalysis.ai)
Column guide
- No.
- Display order only. Sorting changes this number; it is not a published capability rank.
- Model
- The evaluated model name and its provider icon. A dash in a score column means no published score is available here, not a zero.
- DeepSWE v1.1DeepSWE software engineering v1.1
- Tests long-horizon software engineering on original tasks, with success checked against task requirements. Compare the same benchmark version and agent setup. Evaluation methodology ↗
- FrontierSWE v2FrontierSWE software engineering v2
- Tests difficult engineering and technical research tasks. Scores reflect the published task evaluation, with version and agent setup affecting comparability. Evaluation methodology ↗
- Terminal-bench 4.0Terminal benchmark 4.0
- Tests completing practical tasks in a terminal environment, such as coding and system operations. Versions use different task sets and should be compared separately. Evaluation methodology ↗
- Vibe Code BenchVibe coding benchmark
- Tests building web applications from natural-language requests, evaluating the resulting application rather than isolated code answers. Evaluation methodology ↗
Read scores within the same benchmark, source and version. Higher values indicate stronger performance on the reported metric, but different columns do not share a common scale.