What happened
The LongCat team published LongCat-2.5 on its blog on November 1, 2026, describing it as a native multimodal model for long-horizon agentic tasks. The blog says long-context training and W4A4 inference were optimized on domestic AI accelerators. It has 1.6 trillion total parameters with about 48 billion activated per token and builds on LongCat-2.0 with stronger visual understanding and agentic coding.
The platform change log for November 2 lists LongCat-2.5 as available, with three highlights: native multimodality for visual coding, creative design and 3D tasks; long-horizon agents for complex codebases and full-project builds; and compatibility with Claude Code, Codex, Hermes and other agent harnesses.
What the developer reports
On Terminal Bench v4.0, LongCat reports a score of 38.5, measured in the Claude Code harness. LongCat's own scores are self-reported; rival scores come partly from official posts and leaderboards and partly from LongCat's own runs, and the main comparison charts set it against GPT-5.5, GPT-5.6-Sol, Claude Opus 4.8 and Claude Opus 5, not the newest frontier models.
The blog and change log we checked do not state LongCat-2.5's context window or output limit. The blog lists GitHub and Hugging Face links as coming soon, so no weights are published yet.
What it means for you
If you run Claude Code, Codex or Hermes, LongCat-2.5 can be plugged in as the model behind the harness through LongCat's API. Start with a task whose result you already know, such as a closed issue with tests, and compare the result and token use with your current model.
Treat the benchmark numbers as vendor claims until independent results appear, and check LongCat's pricing page for current rates before running long agent sessions.
Go to the source
Sources & further reading
Based on the official documentation below. Features and access can change; check the provider before making a decision.
Was this page useful?
Help us make our coverage more useful. This is feedback on the page, not a rating of the tool.
Go a little further