EAGLE-4 Open-Source Speculative Decoding Hits 961.6 Tokens/s

Eagle AI Corp released EAGLE-4, an open-source speculative decoding method reaching 961.6 tokens/s on Qwen3.8-27B. Available on GitHub as eagleharness, it integrates with SGLang, vLLM, and TokenSpeed via engine plugins.

2026-10-21 ~ 2026-10-21 · 2 related posts

• EAGLE-4 speculative decoding hits 961.6 tokens/s on Qwen3.8-27B, code open-sourced — zhyncs42 · 2026-10-21
• EAGLE-4 speculative decoding goes open source with SGLang and vLLM plugins — testingcatalog · 2026-10-21