{"name":"MegaCapybara","repo":"perkel666/MegaCapybara","url":"https://midorreal.com/project/perkel666-megacapybara","repository":"https://github.com/perkel666/MegaCapybara","description":"The fastest inference engine for Qwen3.8-27B on the NVIDIA RTX 5090: up to 500 tokens/s for one agent and up to 2,000 tokens/s for many, contexts up to 1M tokens, and a launcher that shows what every setting costs. Windows and Linux.","facts":{"license":{"value":"MIT","source":"https://github.com/perkel666/MegaCapybara","checked":"21 Nov 2026"},"homepage":{"value":"https://huggingface.co/perkel/Qwen3.8-27B-MC","source":"https://huggingface.co/perkel/Qwen3.8-27B-MC","checked":"21 Nov 2026"},"last_release":{"value":"v1.40, 4 Nov 2026","source":"https://github.com/perkel666/MegaCapybara/releases","checked":"21 Nov 2026"},"releases":{"value":"12","source":"https://github.com/perkel666/MegaCapybara/releases","checked":"21 Nov 2026"},"contributors":{"value":"1","source":"https://github.com/perkel666/MegaCapybara/graphs/contributors","checked":"21 Nov 2026"},"archived":{"value":"no","source":"https://github.com/perkel666/MegaCapybara","checked":"21 Nov 2026"}},"stars":85.0,"writeup":{"what_it_is":"MegaCapybara is an inference engine for Qwen3.8-27B and Qwen3.8 Flash-Next on the NVIDIA GeForce RTX 5090, running on Windows 11 and Linux. It includes a launcher that shows the accuracy, speed and memory cost of each setting before loading the model, and it serves OpenAI- and Anthropic-compatible APIs locally.","audience":"Users running Qwen3.8 models locally on an RTX 5090 for coding agents","claims":[{"kind":"specific","claim":"Runs Qwen3.8-27B with over 500 tokens/s for one coding agent on the RTX 5090","excerpt":"Over 500 tokens/s for one coding agent","status":"not_checked"},{"kind":"specific","claim":"Speaks both OpenAI's and Anthropic's APIs, including tool calls, thinking and images","excerpt":"MegaCapybara speaks both OpenAI's and Anthropic's APIs, including tool calls, thinking and images","status":"not_checked"},{"kind":"specific","claim":"Requires an NVIDIA GeForce RTX 5090 with driver R580 or newer","excerpt":"An **NVIDIA GeForce RTX 5090** with driver **R580** or newer.","status":"not_checked"},{"kind":"broad","claim":"Is state-of-the-art inference for Qwen3.8-27B","excerpt":"State-of-the-art inference for Qwen3.8-27B on the RTX 5090","status":"too_broad"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-12"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":23,"reality":37,"gap":-13,"momentum":null,"confidence":60,"version":"score-2.0","rule":"verdict-2.0","calculated_at":"2026-11-22T08:00:00+00:00","verdict_source":"rule"},"early_read":null,"changes":[{"date":"2026-11-04T13:34:20+00:00","text":"Latest release v1.40","source":"https://github.com/perkel666/MegaCapybara/releases"},{"date":"2026-10-03T00:29:48+00:00","text":"First release","source":"https://github.com/perkel666/MegaCapybara/releases"},{"date":"2026-10-03T00:29:30+00:00","text":"Repository created","source":"https://github.com/perkel666/MegaCapybara"}],"cite":"MegaCapybara is a GitHub project at https://github.com/perkel666/MegaCapybara, described there as \"The fastest inference engine for Qwen3.8-27B on the NVIDIA RTX 5090: up to 500 tokens/s for one agent and up to 2,000 tokens/s for many, contexts up to 1M tokens, and a launcher that shows what every setting costs. Windows and Linux\".","attribution":"Data from Mid or Real, https://midorreal.com/project/perkel666-megacapybara, checked 22 November 2026. Attribution with a link is required.","still_real":null}