Models Are Getting Really Good At Git

GitBench’s latest evaluation of AI models performing Git-related tasks finds substantial gains across both frontier and open-weight systems, with some newer releases outperforming expectations on accuracy and cost. Fable improved from roughly an 82% to 95% pass rate for text output, while its JSON performance also rose and no longer declined under increased reasoning effort; Opus 5.5 likewise reversed a previous downward trend to rank near the top. OpenAI’s Luna improved its score while becoming less expensive despite using more time and tokens, and Meta’s Muse 1.3 strengthened the company’s standing after weaker recent Llama releases. Open models also showed notable progress, particularly GLM 5.3 Flash and DeepSeek 4.1 Flash, which substantially exceeded their predecessors while retaining low costs. As model performance increasingly approaches the limits of the current benchmark fixtures, the author expects GitBench will need more difficult tests, potentially involving larger contexts, more complex repositories, or an execution harness.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.