HOT RADAR
agentHEAT 4.8°

AI CURATED HOT STORY · 11/20/2026

Claude Writes Its Own Inference Engine, Cutting First-Token Latency to 0.48 Seconds

1 reports archived1 independent sourcesupdated 11/20/2026, 00:18:00
Synthesis & Latest Updates
1 Sources Cross-Validated

Researchers from CMU, NVIDIA and other institutions posed a task in the Singularity-Bench preprint: give Claude Opus 5.5 base server code, four GB300 GPUs and 12 hours, with no internet access, to build an inference engine using Claude Code. Replaying 16 concurrent real coding sessions, the agent-written engine reached roughly 0.48s P90 time-to-first-token versus about 12s for the mature engine SGLang v0.5.20, moving the question of whether agents can build agents from speculation to reproducible engineering evidence.

LATEST/Researchers from CMU, NVIDIA and other institutions posed a task in the Singularity-Bench preprint: give Claude Opus 5.5 base server code, four GB300 GPUs and 12 hours, with no internet access, to build an inference engine using Claude Code. Replaying 16 concurrent real coding sessions, the agent-written engine reached roughly 0.48s P90 time-to-first-token versus about 12s for the mature engine SGLang v0.5.20, moving the question of whether agents can build agents from speculation to reproducible engineering evidence.

Hourly heat curve

18 fully observed hours
Now
4.8
Peak
7.123:00
24h change
-32%

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

Coverage timeline

Total 1 reports · Latest first
  1. 新智元 (微信公众号)T2·78 pts
    • Claude Opus 5.5 independently built an inference engine with Claude Code under no-internet, 12-hour, four-GPU constraints
    • In 16-way concurrent coding-session replay, P90 time-to-first-token fell from about 12s on SGLang to roughly 0.48s
    • Singularity-Bench turns 'can agents build agents' into a quantifiable, reproducible benchmark task