{"name":"EAGLE-4","repo":"eagle-hq/EAGLE-4","url":"https://midorreal.com/project/eagle-hq-eagle-4","repository":"https://github.com/eagle-hq/EAGLE-4","description":"Official Inference Code of EAGLE-4","facts":{"language":{"value":"Python","source":"https://github.com/eagle-hq/EAGLE-4","checked":"22 Nov 2026"},"homepage":{"value":"https://eagle.ai","source":"https://eagle.ai","checked":"22 Nov 2026"},"last_release":{"value":"v0.1.5, 1 Nov 2026","source":"https://github.com/eagle-hq/EAGLE-4/releases","checked":"22 Nov 2026"},"releases":{"value":"6","source":"https://github.com/eagle-hq/EAGLE-4/releases","checked":"20 Nov 2026"},"contributors":{"value":"2","source":"https://github.com/eagle-hq/EAGLE-4/graphs/contributors","checked":"20 Nov 2026"},"archived":{"value":"no","source":"https://github.com/eagle-hq/EAGLE-4","checked":"22 Nov 2026"}},"stars":1222.0,"writeup":{"what_it_is":"EAGLE-4 is an inference speedup project for large language models using speculative decoding. This repository holds eagle-harness, a package that connects EAGLE-4 draft models to the SGLang, vLLM and TokenSpeed engines, and it adjusts how far to speculate based on measured costs.","audience":"Engineers serving large language models with SGLang, vLLM or TokenSpeed on GPUs","claims":[{"kind":"specific","claim":"Speeds up generation with speculative decoding, up to 3.4x on Qwen3.8-27B","excerpt":"up to 3.4× faster generation on Qwen3.8-27B, with the same answers","status":"not_checked"},{"kind":"specific","claim":"Decides how far to speculate from measured costs on the user's setup","excerpt":"decides how far to speculate from measured costs on your setup","status":"not_checked"},{"kind":"specific","claim":"Keeps pace with plain decoding up to 512 concurrent requests on four B300 GPUs with the high-load recipe","excerpt":"keeps pace with plain decoding up to 512 concurrent requests","status":"not_checked"},{"kind":"broad","claim":"Output stays exactly what the target model would generate on its own","excerpt":"The output stays exactly what the target model would generate on its own","status":"too_broad"}],"alternatives":[],"written_by":"AI, from the project's README","date":"2026-11-05"},"why_now":null,"owner_supplied":null,"score":{"verdict":"mid","hype":26,"reality":52,"gap":-26,"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":"2026-11-01T02:14:00+00:00","text":"Latest release v0.1.5","source":"https://github.com/eagle-hq/EAGLE-4/releases"},{"date":"2026-10-20T17:41:49+00:00","text":"First release","source":"https://github.com/eagle-hq/EAGLE-4/releases"},{"date":"2026-10-20T14:03:04+00:00","text":"Repository created","source":"https://github.com/eagle-hq/EAGLE-4"}],"cite":"EAGLE-4 is a GitHub project at https://github.com/eagle-hq/EAGLE-4 written mainly in Python, described there as \"Official Inference Code of EAGLE-4\".","attribution":"Data from Mid or Real, https://midorreal.com/project/eagle-hq-eagle-4, checked 22 November 2026. Attribution with a link is required.","still_real":null}