Better engineering, measured
Descant brings multiple frontier models together in a complete software engineering system. On DeepSWE v1.1, Descant Soprano 7 resolved 93 of 113 tasks, scoring 82.3%.
- Tasks resolved
- 93 of 113
- Descant system release
- Soprano 7
How the result compares
Descant’s system-level result is shown alongside the published frontier-model scores below.
| System or model | DeepSWE v1.1 |
|---|---|
| Descant · Soprano 7 | 82.3% |
| Muse Wonder | 79.0% |
| Gemini 4 Argon | 77.9% |
| GPT-6.1 Sol | 75.2% |
| Claude Opus 5.5 | 74.2% |
| GPT-6 Astra | 74.1% |
| Grok 4.7 | 71.0% |
| Claude Fable 5 | 70.0% |
Comparison published with Descant’s October 14, 2026 report. Frontier-model scores sourced from llm-stats.com as of October 14, 2026.
What the benchmark measures
DeepSWE uses original, long-horizon engineering tasks in existing codebases. It tests whether a system can understand the project and implement a substantial change.
The result measures Descant as a complete system, including the engineering work required to turn an assignment into a solution.
Continuous development adds a longer horizon: coordinating related issues, retaining project knowledge and advancing a codebase through many changes. A single-task benchmark does not measure that entire capability.
Read the evaluation reportLook beyond the score
The benchmark work includes 113 task repositories, solutions, pull requests and task notes. Access is available on request if you want to examine how Descant approached the work.
See what Descant can do with your work
Start with an issue in your own codebase, or bring us an initiative you want to discuss. Evaluate the engineering, the progress and the cost against the work that matters to your team.