Grok 4.5 has taken the lead on the SWE Marathon benchmark, scoring 29.0% and beating two well-known models: Claude Opus 4.8 and Fable. The result places the latest version of Grok at the front of a ranking that tests how well AI systems handle real-world software engineering tasks. The model's API is priced at $2 per million tokens.
What the benchmark shows
The SWE Marathon benchmark is designed to measure a model's ability to solve software engineering problems. Grok 4.5's 29.0% score puts it ahead of Claude Opus 4.8 and Fable, though the exact margins between them were not disclosed. The benchmark's focus on practical coding tasks makes it a useful gauge for developers who want to evaluate AI assistants for actual development work.
Leading a benchmark doesn't guarantee a model will be the best choice for every project. But the result signals that Grok 4.5 has improved its software engineering chops compared to earlier versions or competitors. The test covers a broad set of challenges, from bug fixes to feature implementation, so a top score suggests strong overall capability.
API pricing and access
Alongside the benchmark news, the team behind Grok 4.5 set API pricing at $2 per million tokens. That's a straightforward rate for developers looking to integrate the model into their own tools and workflows. The pricing is competitive with other large language models, though the exact features and rate limits tied to that tier haven't been detailed.
For teams that need to process large volumes of code or documentation, the per-token cost will be a key factor. At $2 per million tokens, a typical software engineering query might cost fractions of a cent. Builders can now test the model against their own benchmarks to see if the SWE Marathon top spot translates to better results on their specific tasks.
Developers interested in trying Grok 4.5 can access it through the API at that rate. The model's performance on the SWE Marathon benchmark gives them a concrete data point to weigh against other options on the market.
