EAGLE-4 is an open-source speculative-decoding system with one trained drafter per target model that adapts draft-tree depth, width and verification budget to hardware and batch size. It reports about 4x faster decoding on Qwen3.8-27B, with SGLang and vLLM integrations for NVIDIA B300 and AMD MI355X.
Read the official announcement ↗Home / New releases / October 2026 / Week of Oct 19–25, 2026
⚡ EAGLE-4 speculative decoding
Released October 20, 2026 · Verified against the official source · Official announcement (blog.eagle.ai) ↗
Also released the week of Oct 19–25, 2026
- 2026-10-25 Pomerado
- 2026-10-25 Nanbeige4.3-33B (Nanbeige LLM Lab)
- 2026-10-23 Nemotron 3.5 Bolt E4B 4GB (NVIDIA)
- 2026-10-23 MiniMax M3.1 Flash open weights
- 2026-10-23 Kimi K3.1 Pro + Highspeed (Moonshot AI)
- 2026-10-23 Hy-Nomas interactive world model (Tencent Hunyuan)
- 2026-10-23 Hi3D V1.0 open source (formerly Sparc3D)
- 2026-10-22 Sume API (Sume Labs)
Frequently asked questions
When was EAGLE-4 speculative decoding released?
EAGLE-4 speculative decoding was released on October 20, 2026, according to the official announcement (blog.eagle.ai). SIloadout added it to its verified release archive for the week of Oct 19–25, 2026.
Where is the official announcement for EAGLE-4 speculative decoding?
The official source is https://blog.eagle.ai/blog/introducing-eagle-4. EAGLE-4 is an open-source speculative-decoding system with one trained drafter per target model that adapts draft-tree depth, width and verification budget to hardware and batch size.
← Newer: Eleven Realtime (ElevenLabs) · Older: Beacon (Asymptote Labs) → · Week of Oct 19–25, 2026 · October 2026 · All releases · JSON Feed