Celeris-1 claims 2,038 tokens per second and fastest-on-leaderboard status
Scobleizer · x · 2026-08-04
- Scobleizer reposts Celeris AI’s claim that Celeris-1 is now the fastest AI model on the Artificial Analysis leaderboard.
- The post cites 2,038 output tokens per second, claiming #1 across 591 models on response time, TTFT, and time-per-intelligence-task.
- It also says the model is 2× faster than the next fastest system and beats Grok 4.3 on HLE.
- The key claim is that this level of speed is possible without custom silicon.
More from Models
- Meituan’s LongCat 2.0 is praised for domestic chips but judged mid-pack on quality — 葬AI · 2026-08-04
- Google lets Gemini 3.5 Flash and 3.6 Flash use Maps and Search together — OfficialLoganK · 2026-08-04
- DeepSeek V4 Flash tops the Vals Index above 60 at 35x lower cost — zephyr_z9 · 2026-08-04
- gpt-oss-20b beats Opus on a task while being 1,000x cheaper and faster — scaling01 · 2026-08-04
- Vibe Code Bench shows V4 Flash beating GLM 5.2 on score and price — teortaxesTex · 2026-08-04
- Claude aced advanced math tutoring, then botched a flight-speed table — Recent-Day3062 · 2026-08-04