FrontierCode 1.1
An evaluation benchmark for AI coding capabilities that simulates real-world software engineering tasks, covering scenarios such as code completion, bug fixing, multi-file refactoring, and unit test generation. Test cases are drawn from real open-source projects and require models to demonstrate cross-file context understanding and long-sequence reasoning.
Core Facts
FrontierCode version 1.1 introduces more complex multi-file project operations, cross-language code understanding, and debugging tasks requiring deep reasoning compared to earlier versions
FrontierCode 1.1's test cases are typically derived from real open-source projects and require models to have cross-file context understanding and long-sequence reasoning capabilities.
FrontierCode 1.1 is an evaluation benchmark for AI programming capabilities that simulates real software engineering tasks, including scenarios such as code completion, bug fixing, multi-file refactoring, and unit test generation.
Timeline (last 90 days)
Gemini 4 Argon and GPT-6 Sol tie on FrontierCode 1.1 and beat GPT-6.1 Sol, indicating that the gap in coding ability among current top models has narrowed to a very small range.
Gemini 4 Argon's score on FrontierCode 1.1 is on par with GPT-6 Sol.
Gemini 4 Argon achieved a score of 60.5% on the code competition benchmark FrontierCode 1.1.
Compared with the previous generation, FrontierCode version 1.1 strengthens multilingual coverage and long-context-dependent question types.
Claude Haiku 5.5 scored 58.4% on the FrontierCode 1.1 benchmark.
Under the low effort setting, GPT-6.1 Sol achieves a score of 58.1% on FrontierCode 1.1, with a per-task cost of only $0.21.
On the FrontierCode 1.1 benchmark, Sonnet 5.5 achieved notable improvement compared to its predecessor Sonnet 5.
FrontierCode 1.1's task cost metric incorporates model invocation expenses into the evaluation dimension, enabling performance and economic efficiency to be compared horizontally within the same coordinate system.
Claude Opus 5.5 topped the FrontierCode 1.1 benchmark, surpassing Fable 5 and GPT-6 Astra.
On the overall aggregate score of FrontierCode 1.1, Grok 4.7 slightly trails its predecessor Grok 4.6.
7 more timeline events
All Facts (16)
FrontierCode version 1.1 introduces more complex multi-file project operations, cross-language code understanding, and debugging tasks requiring deep reasoning compared to earlier versions
70%VerifiedFrontierCode 1.1's test cases are typically derived from real open-source projects and require models to have cross-file context understanding and long-sequence reasoning capabilities.
65%VerifiedFrontierCode 1.1 is an evaluation benchmark for AI programming capabilities that simulates real software engineering tasks, including scenarios such as code completion, bug fixing, multi-file refactoring, and unit test generation.
65%VerifiedFrontierCode is an evaluation benchmark for AI models' programming capabilities focusing on real engineering tasks rather than simple algorithm problems
65%UnverifiedClaude Opus 5 approaches Fable-level performance on the FrontierCode 1.1 benchmark while costing only half as much
60%UnverifiedGemini 4 Argon's score on FrontierCode 1.1 is on par with GPT-6 Sol.
50%UnverifiedGemini 4 Argon and GPT-6 Sol tie on FrontierCode 1.1 and beat GPT-6.1 Sol, indicating that the gap in coding ability among current top models has narrowed to a very small range.
50%UnverifiedGemini 4 Argon achieved a score of 60.5% on the code competition benchmark FrontierCode 1.1.
50%UnverifiedClaude Haiku 5.5 scored 58.4% on the FrontierCode 1.1 benchmark.
50%UnverifiedCompared with the previous generation, FrontierCode version 1.1 strengthens multilingual coverage and long-context-dependent question types.
50%UnverifiedUnder the low effort setting, GPT-6.1 Sol achieves a score of 58.1% on FrontierCode 1.1, with a per-task cost of only $0.21.
50%UnverifiedOn the FrontierCode 1.1 benchmark, Sonnet 5.5 achieved notable improvement compared to its predecessor Sonnet 5.
50%UnverifiedFrontierCode 1.1's task cost metric incorporates model invocation expenses into the evaluation dimension, enabling performance and economic efficiency to be compared horizontally within the same coordinate system.
50%UnverifiedOn the overall aggregate score of FrontierCode 1.1, Grok 4.7 slightly trails its predecessor Grok 4.6.
50%UnverifiedIn the FrontierCode 1.1 benchmark, Grok 4.7 performs strongest on difficult, multi-module backend tasks.
50%Unverified在 Cognition 推出的第三方基准 Frontier Code 上，Sonnet 5 超越了 Opus 4.8
50%