A coding model becomes useful when it can follow a task through several steps and leave behind work that can be checked. LongCat-2.5’s API release puts that workflow at the center. Its documentation now gives developers enough detail to plan a bounded trial, without treating a product claim as a completed project.
Read the dated primary source ↓What the November 2 release establishes
The official LongCat API changelog lists LongCat-2.5 as available on November 2, 2026. It describes native multimodality for visual coding, creative design and 3D tasks, alongside multi-step reasoning and tool use for complex codebases and full-project builds.
The same entry names Claude Code, Codex and Hermes among supported agent harnesses. That is a compatibility claim from the provider, not evidence that every existing workflow will behave identically after switching models.
The changelog separately dates LongCat-2.5-Preview to September 25. This story concerns the new API availability entry; it should not be read as saying the model family appeared for the first time yesterday.
Public record: LongCat API changelog — November 2, 2026
API access and the documented limits
The quick-start guide lists LongCat-2.5 as the model name and documents OpenAI-format Chat Completions and Anthropic-format Messages access. Getting started requires a platform account and an API key.
The guide specifies a 1M-token context window and maximum output length of 128K tokens. For rate-limit responses, it documents HTTP 429 and recommends exponential backoff.
A large context window can help a workflow carry more material, but it does not establish that the model will correctly use every relevant file. Test whether it finds the right dependency, preserves existing behavior and supplies evidence for its proposed change. Those are separate questions from how much input a request accepts.
Public record: LongCat API quick start — checked November 3, 2026
Introductory rates are a starting point for budgeting
LongCat’s pricing page lists limited-time discounted rates of $0.30 per million uncached input tokens, $0.006 per million cached input tokens and $1.20 per million output tokens. The page says prices may change and that platform prices and billing records are authoritative. It does not show an end date for the discount.
Using those published rates, a request with 100,000 uncached input tokens and 10,000 output tokens would cost $0.042 in model-token charges. This is an illustrative calculation, not a forecast for an agent session. Repeated attempts and further model calls add to the total.
For a useful cost comparison, record tokens and retries for the complete task. A cheap request that needs several repairs can have a different cost from a successful first attempt.
Public record: LongCat-2.5 pricing — checked November 3, 2026
A first trial should end in an inspectable change
Our suggested test is one small repository task with a clear acceptance condition: reproduce a bug, make the smallest appropriate repair and run a check that would fail without that repair. Keep the task, tools and time budget consistent when comparing models.
For the visual side, provide a permitted interface reference and ask for one bounded implementation change. Compare the result against the reference at the intended screen size, then check the behavior separately. Visual resemblance and functioning controls need their own review.
Promptwire has not independently tested LongCat-2.5. The release makes it a candidate for evaluation; the decision to adopt it should rest on accepted changes, reviewer effort and total task cost in the workflow you actually use.
Follow the evidence.
Primary documentation checked November 3, 2026. Release capabilities and compatibility are provider claims. Promptwire has not independently benchmarked the model. The cost example is calculated from the listed rates; the trial is an editorial recommendation. Cover: AI-assisted conceptual illustration, not a product screenshot.
- LongCat API changelog — November 2, 2026
- LongCat API quick start — checked November 3, 2026
- LongCat-2.5 pricing — checked November 3, 2026
Public sources support the factual background. Analysis and illustrative exercises are identified in the text. How we assess AI · Editorial standards
Spot a factual error or an outdated detail?
