As of the collection date, the LongCat-2.5-Preview and OpenCode combination scored 44.65/70 on the site's coding leaderboard, ranking 44th. This is a snapshot of a specific task and tool combination.
Tasks this result can help assess: Short coding tasks based on existing codebases, including code quality, bug finding, and edge-case testing.
Tasks this result should not be generalized to: Multimodal understanding, long-running autonomous development, multi-agent collaboration, and performance with other clients.
Applicable model version: LongCat-2.5-Preview.
Test environment or client: OpenCode; the model provider, specific version, and runtime environment were not specified.
Reasoning level and parameters: Not specified.
The leaderboard's methodology says the test covers 7 existing projects and 27 prompts across Laravel/PHP, React/TypeScript, Flutter/Dart, and Go. Each run starts from a fresh copy of a project, with no retries or human guidance mid-run; each prompt is run multiple times. This methodology page reflects the current explanation visible at the time of collection; it cannot be confirmed that every detail was identical during the October 2 test. The title of the video associated with the leaderboard mentions 24 coding prompts, which differs from the current methodology page. It should not be treated as definitive evidence of the prompt count for the same test.
The model-and-tool combination scored 44.65/70 overall. Code quality scores for Laravel and React-TS were 16.98/20 and 16.67/20, respectively; bug finding scored 2.4/10; edge-case testing across four projects scored 10.6/20; and the consistency penalty was 2.00 points.
| Metric | Value on page |
|---|---|
| Leaderboard rank | 44 (2026-11-03 snapshot) |
| Total score | 44.65/70 |
| Laravel code quality | 16.98/20 |
| React-TS code quality | 16.67/20 |
| Bug finding | 2.4/10 |
| Edge-case testing | 10.6/20 |
| Consistency penalty | −2.00 |
| Average time per prompt | 16 minutes 50 seconds |
| Average cost per prompt | N/A |
The page says the cost for tests under the subscription plan is listed as N/A; this is not enough to calculate the model's API cost.
The score measures the model-and-tool combination, not the model's capabilities in isolation; the leaderboard ranking may change as new tests are added. The page does not provide the number of individual runs for this model, reasoning parameters, or confidence intervals, so the statistical significance of the differences cannot be assessed. The associated video and current methodology page differ in their prompt counts; the original video and test records would need to be checked to confirm the details of that run.
See the original leaderboard page, test methodology, and associated video. The page does not disclose enough specific configuration to guarantee strict reproduction of this score.
LongCat 2.5
