LongCat 2.5 Free

Meituan's LongCat team introduced LongCat 2.5 as a natively multimodal mixture-of-experts model with 1.6 trillion total parameters and about 48 billion active per token. The technical blog appeared on November 1, 2026, followed by general API availability on November 2, with the model built for long-horizon coding-agent work. The supplied write-up reports a 1,048,576-token context window, text, image, and video input with text output, and up to 262,144 output tokens. It also lists Anthropic- and OpenAI-format endpoints and setup support for Claude Code, Codex, Hermes Agent, OpenCode, and Cline, while noting that 2.5 weights remain forthcoming.

LongCat 2.5 is listed on LongCat's international pay-as-you-go API at $0.30 per million uncached input tokens, $0.006 per million cached input tokens, and $1.20 per million output tokens, according to a Tabbit Browser guide whose rates were checked on November 3, 2026. The guide notes these are limited-time discounted rates and that the provider has not published an end date. The guide argues cached input effectively costs 98% less than uncached input, but warns the discount applies only to tokens actually reported as cache hits, not to output or the whole task. It also cautions that identical displayed rates across LongCat 2.0 and 2.5 do not imply identical task cost, and that 2.0's published weights and provider routes should not be assumed to apply to 2.5.

OpenCode's Go subscription documentation lists LongCat 2.5 Free among the models available to Go, Go Plus, and Go Teams subscribers, marked as limited time. The plan tiers start at $10/month for Go, $40/month for Go Plus, and $20 per member per month for Go Teams, with each member receiving its own limits. The same page shows that access to LongCat 2.5 Free is region-gated by workspace privacy settings, and owners or admins can enable Global under Regions in Settings › Privacy to open it up. By default, paid endpoints that train on request data are blocked, with allowed regions set to Europe and United States.