FrontierCode 1.1 Extended
We show this table for reference; we do not rank on it.
Cognition's 150-task Extended subset of the FrontierCode 1.1 software-engineering benchmark.
Current release
FrontierCode 2.0 is the current release. Its 132 tasks are larger, come from more repositories, and add stricter blocking criteria, especially for thoroughness, so its scores are not directly comparable with FrontierCode 1.1.
View FrontierCode 2.0Extended score on FrontierCode 1.1 Extended — November 12, 2026 snapshot
We mirror the published extended score view for FrontierCode 1.1 Extended. Claude Opus 5.5 leads the public snapshot at 65.3%, followed by Claude Fable 5 (64.9%) and GPT-6 Astra (64.5%). We do not use these results to rank models overall.
Claude Opus 5.5
Anthropic
claude-code · medium reasoning
Claude Fable 5
Anthropic
claude-code · xhigh reasoning
GPT-6 Astra
OpenAI
codex · max reasoning
45 modelsCodingCurrentDisplay onlyUpdated November 12, 2026 snapshot
Extended score table (45 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Opus 5.5Anthropicclaude-code · medium reasoning | 65.3% | Not reported | Closed |
| 2 | Claude Fable 5Anthropicclaude-code · xhigh reasoning | 64.9% | Not reported | Closed |
| 3 | GPT-6 AstraOpenAIcodex · max reasoning | 64.5% | Not reported | Closed |
| 4 | Claude Sonnet 5.5Anthropicclaude-code · xhigh reasoning | 64.4% | Not reported | Closed |
| 5 | Claude Fable 5.1Anthropicclaude-code · medium reasoning | 63.6% | Not reported | Closed |
| 6 | Claude Opus 5Anthropicclaude-code · medium reasoning | 63.6% | Not reported | Closed |
| 7 | SWE-2Cognitiondevin · max reasoning | 62.5% | Not reported | Closed |
| 8 | Grok 4.6xAIgrok-build · high reasoning | 61.3% | Not reported | Closed |
| 9 | GPT-6 SolOpenAIcodex · max reasoning | 60.7% | Not reported | Closed |
| 10 | Gemini 4 ArgonGoogledevin-cli · low reasoning | 60.6% | Not reported | Closed |
| 11 | GPT-5.6 SolOpenAIcodex · max reasoning | 60.6% | Not reported | Closed |
| 12 | GPT-6.1 SolOpenAIcodex · medium reasoning | 60.4% | Not reported | Closed |
| 13 | Claude Opus 4.8Anthropicclaude-code · max reasoning | 59.6% | Not reported | Closed |
| 14 | Grok 4.7xAIgrok-build · high reasoning | 59.4% | Not reported | Closed |
| 15 | Claude Haiku 5.5Anthropicclaude-code · max reasoning | 58.4% | Not reported | Closed |
| 16 | Kimi K3Moonshot AImini-swe-agent · none reasoning | 58.2% | Not reported | Open |
| 17 | GPT-5.5OpenAIcodex · xhigh reasoning | 56.7% | Not reported | Closed |
| 18 | DeepSeek V4.1 Flash 1017DeepSeekdevin-cli · high reasoning | 56.6% | Not reported | Open |
| 19 | Grok 4.5xAIgrok-build · high reasoning | 56.5% | Not reported | Closed |
| 20 | Gemini 3.7 FlashGoogledevin-cli · medium reasoning | 56.3% | Not reported | Closed |
| 21 | Claude Sonnet 5Anthropicclaude-code · xhigh reasoning | 56.2% | Not reported | Closed |
| 22 | GPT-6 LunaOpenAIcodex · max reasoning | 56.1% | Not reported | Closed |
| 23 | GPT-5.6 TerraOpenAIcodex · max reasoning | 55.8% | Not reported | Closed |
| 24 | GPT-5.6 LunaOpenAIcodex · max reasoning | 55.1% | Not reported | Closed |
| 25 | SWE-1.7Cognitiondevin-cli · none reasoning | 54.3% | Not reported | Closed |
| 26 | Claude Opus 4.7Anthropicclaude-code · max reasoning | 53.9% | Not reported | Closed |
| 27 | Gemini 3.8 FlashGoogledevin-cli · medium reasoning | 53.4% | Not reported | Closed |
| 28 | GLM-5.3Z.AIdevin-cli · max reasoning | 52.6% | Not reported | Open |
| 29 | Gemini 3.6 FlashGoogledevin-cli · medium reasoning | 48.0% | Not reported | Closed |
| 30 | GLM-5.3-FlashZ.AIdevin-cli · max reasoning | 46.2% | Not reported | Open |
| 31 | Kimi K2.7Moonshot AImini-swe-agent · none reasoning | 45.4% | Not reported | Unknown |
| 32 | DeepSeek V4 Pro 0813DeepSeekdevin-cli · high reasoning | 44.5% | Not reported | Open |
| 33 | Claude Opus 4.6Anthropicclaude-code · high reasoning | 43.7% | Not reported | Closed |
| 34 | GPT-5.4 miniOpenAIcodex · xhigh reasoning | 43.0% | Not reported | Closed |
| 35 | Composer 2.5Cursorcursor-cli · none reasoning | 40.8% | Not reported | Closed |
| 36 | GLM-5.2Z.AImini-swe-agent · none reasoning | 40.1% | Not reported | Open |
| 37 | Claude Sonnet 4.6Anthropicclaude-code · max reasoning | 40.0% | Not reported | Closed |
| 38 | DeepSeek V4 Flash 0731DeepSeekdevin-cli · high reasoning | 31.7% | Not reported | Open |
| 39 | DeepSeek V4 ProDeepSeekmini-swe-agent · none reasoning | 31.0% | Not reported | Unknown |
| 40 | MiniMax M3MiniMaxmini-swe-agent · none reasoning | 28.5% | Not reported | Open |
| 41 | Nemotron 3 UltraNVIDIAdevin-cli · none reasoning | 27.2% | Not reported | Open |
| 42 | InklingThinking Machines Labmini-swe-agent · 0.99 reasoning | 24.8% | Not reported | Open |
| 43 | Qwen3.7 PlusAlibabamini-swe-agent · none reasoning | 21.8% | Not reported | Closed |
| 44 | SWE-1.6Cognitiondevin-cli · none reasoning | 20.5% | Not reported | Unknown |
| 45 | Mistral 3.5 MediumMistraldevin-cli · none reasoning | 18.7% | Not reported | Unknown |
How we show FrontierCode 1.1 Extended
The snapshot mirrors Cognition's FrontierCode 1.1 Extended table captured on November 12, 2026 snapshot. Main uses the 100 hardest tasks from the 150-task Extended set. Each table selects the best published reasoning effort for its own task set.
We keep these mirrored results display-only. Each row combines a model with an agent harness, and the tasks are private, so the mirror does not enter weighted model rankings. Runs flagged for unfair internet use receive zero in the published score.
Snapshot
The published FrontierCode 1.1 Extended snapshot places Claude Opus 5.5 first at 65.3%. The third row is 0.8 points behind. The broader top-10 range is 4.7 points, so many of the published results sit in a relatively narrow band.
45 models have been evaluated on FrontierCode 1.1 Extended. The benchmark falls in the Coding category. FrontierCode 1.1 Extended is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About FrontierCode 1.1 Extended
Year
2026
Tasks
150 private software-engineering tasks
Format
Repository task completion with maintainer rubrics
Difficulty
Frontier coding-agent quality
FrontierCode 1.1 Extended contains all 150 private software-engineering tasks. The mirrored leaderboard selects each model-agent row's best published reasoning effort on Extended independently of Main. The mirror is display-only; each result combines a model with an agent harness.
Freshness and provenance
Version
FrontierCode 1.1 Extended
Refresh cadence
Rolling
Staleness state
Current
Question availability
Private tasks with public aggregate results
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does FrontierCode 1.1 Extended measure?
Cognition's 150-task Extended subset of the FrontierCode 1.1 software-engineering benchmark.
Which model leads the published FrontierCode 1.1 Extended snapshot?
Claude Opus 5.5 currently leads the published FrontierCode 1.1 Extended snapshot with 65.3% extended score. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on FrontierCode 1.1 Extended?
The November 12, 2026 snapshot snapshot contains 45 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.