---
type: "post"
id: "eagle-4-ships-as-eagle-harness-a-source-7bdb0f"
slug: "eagle-4-ships-as-eagle-harness-a-source-7bdb0f"
title: "EAGLE-4 ships as eagle-harness , a source-available drafter that plugs into SGLang, vLLM, and TokenSpeed."
description: "EAGLE Research has published an EAGLE-4 draft-model checkpoint for Qwen3.8-27B on Hugging Face."
retrieval_nugget: "EAGLE Research has published an EAGLE-4 draft-model checkpoint for Qwen3.8-27B on Hugging Face."
published_at: "2026-10-21"
updated_at: "2026-10-21"
record_date: "2026-10-21"
date_kind: "published_at"
topics: ["inference","models"]
entities: ["EAGLE Research","EAGLE-4","Qwen3.8-27B","Hugging Face","NVIDIA B300","SPEED-Bench"]
related_patterns: []
novelty_label: "new_fact"
what_changed: "A Hugging Face Qwen3.8-27B EAGLE-4 drafter release adds benchmark results: up to 3.59× throughput at T=0 and 2.59× at T=1 on 4× B300 GPUs."
prior_refs: ["memory:6fb61ca2-d7e7-8d94-a372-a8510b58e525"]
source_url: "https://huggingface.co/EAGLE-corp/EAGLE4-Qwen3.8-27B"
source_urls: ["https://huggingface.co/EAGLE-corp/EAGLE4-Qwen3.8-27B"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:eagle-4-ships-as-eagle-harness-a-source-7bdb0f"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/eagle-4-ships-as-eagle-harness-a-source-7bdb0f/","markdown":"https://newruntime.com/posts/eagle-4-ships-as-eagle-harness-a-source-7bdb0f.md","json":"https://newruntime.com/posts/eagle-4-ships-as-eagle-harness-a-source-7bdb0f.json"}
---

# EAGLE-4 ships as eagle-harness , a source-available drafter that plugs into SGLang, vLLM, and TokenSpeed.

## Retrieval answer

EAGLE Research has published an EAGLE-4 draft-model checkpoint for Qwen3.8-27B on Hugging Face.

EAGLE Research has published an EAGLE-4 draft-model checkpoint for Qwen3.8-27B on Hugging Face. The checkpoint is intended for speculative decoding with a compatible EAGLE-4 implementation and the target model; it is not a standalone text-generation model.

The model card describes EAGLE-4 as a flexible drafter intended to adapt its decoding strategy across batch sizes, hardware platforms, and workloads. In speculative decoding, the draft model proposes candidate tokens which are verified by the target model.

For a qualitative SPEED-Bench split at concurrency one, using four NVIDIA B300 GPUs with tensor parallelism four and a 4,096-token maximum output, the authors report 863.8 tokens per second and a 3.59x speedup over vanilla decoding at temperature zero. At temperature one, they report 608.6 tokens per second and a 2.59x speedup. These are author-reported results under a specified configuration; the source does not establish independent performance, performance on other hardware or workloads, or production availability. It also instructs users to check licenses for both the checkpoint and underlying Qwen3.8-27B weights.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Inference](https://newruntime.com/topics/inference.md): Explore the inference topic hub.
- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)