PulseAugur
中
实时 11:30:44
English(EN) Mercury 3 Does 1,068 Tokens/Second With a Diffusion LLM — Here's the Architecture Trick (and How to Try It)

Inception Labs 发布 Mercury 3 扩散 LLM,速度达 1,068 tokens/秒 · 跟踪到 2 个来源

Inception Labs 推出了 Mercury 3,这是一款基于扩散的 LLM,在 NVIDIA GPU 上实现了每秒 1,068 个 token 的速度,显著优于自回归模型。这种新架构允许跨整个序列并行细化 token,从而加快填充、编辑和约束生成等任务的处理速度。虽然供应商的基准测试声称在质量上与 GPT-6 Luna 和 Claude Haiku 5.5 等模型相当,但独立评估尚待进行,并且扩散 LLM 的生态系统仍在发展中。 AI

影响 通过减少模型推理时间来加速代理任务完成,可能支持新的实时 AI 应用。

排序理由 前沿实验室模型发布,附带系统卡和性能声明。[lever_c_demoted from frontier_release: ic=2 ai=1.0]

在 dev.to — LLM tag 阅读 →

AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →

Inception Labs 发布 Mercury 3 扩散 LLM,速度达 1,068 tokens/秒 · 跟踪到 2 个来源

本文如何被排名

Signal score
18 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Significant
前沿实验室模型发布,附带系统卡和性能声明。[lever_c_demoted from frontier_release: ic=2 ai=1.0]
Source corroboration
2 independent sources
Multiple independent publishers reporting the same story raises confidence that it's real and newsworthy.
Topics
model release, infra
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Same-day
Cluster formed today. Ranking reflects the current source set at time of score.

完整方法见我们的编辑标准。

报道来源 [2]

  1. dev.to — LLM tag TIER_1 English(EN) · Shivam Kumar ·

    Mercury 3 使用扩散式大语言模型实现每秒 1068 个 token — 架构技巧在此(以及如何尝试)

    <p>Today Inception Labs dropped <strong>Mercury 3</strong>: a diffusion-based LLM with a 1M-token context window doing <strong>1,068 tokens/sec on NVIDIA GPUs</strong>, claiming to beat GPT-6 Luna and Claude Haiku 5.5 on benchmarks — at $0.20/$0.70 per million tokens, half price …

  2. dev.to — LLM tag TIER_1 English(EN) · M.T.Ramkrushna ·

    Mercury 3与Diffusion LLM的赌注:每秒1000个Token的改变(以及未改变之处)

    <p>Every step of an AI agent waits on the model.</p> <p>Plan, call a tool, read the result, decide, call another tool. If each model call takes a few seconds, a ten-step task feels like forever, and a voice agent feels broken. Most "make it faster" work today is squeezing the sam…