Celeris-1 ranks as the fastest AI model on Artificial Analysis with over 2,000 output tokens per second.

// 1h agoBENCHMARK RESULT

Celeris-1 ranks as the fastest AI model on Artificial Analysis with over 2,000 output tokens per second.

Celeris AI announced that its Celeris-1 model has taken the top ranking for speed on Artificial Analysis out of 591 evaluated models. Achieving 2,038 output tokens per second, Celeris-1 leads in response time, time to first token (TTFT), and time per intelligence task, delivering twice the generation speed of the next fastest model.

// ANALYSIS

Ultra-low latency and raw inference throughput are becoming major differentiators for next-generation real-time AI applications.

• Generating 2,038 output tokens per second provides a significant speed advantage for real-time voice, interactive agents, and rapid code generation.

• Ranking #1 across 591 models on Artificial Analysis highlights the potential of optimized inference paradigms over standard autoregressive execution.

• Double the speed of the nearest competitor combined with fast TTFT substantially improves user experience for latency-critical workloads.

// TAGS

celeris-1celeris-aiartificial-analysisinference-speedllm-benchmarklatency

DISCOVERED

1h ago

2026-08-04

PUBLISHED

1h ago

2026-08-04

RELEVANCE

9/ 10

AUTHOR

Celeris_ai

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

UPDATE15m ago

Claude Code v2.1.221 adds sandbox mask mode

The v2.1.221 release of Claude Code adds support for mask mode within its sandbox environment on Linux and WSL2. By adding .env files to the sandbox credentials configuration, executing commands and tools automatically read fake values instead of real sensitive secrets.

UPDATE40m ago

OpenCode 1.18.12 eliminates composer lag, sharpens diff reviews

OpenCode version 1.18.12 brings targeted performance and stability improvements to AI-assisted coding workflows by migrating prompt drafts to async IndexedDB with buffering to eliminate composer lag. The release also sharpens diff review precision, resolves Azure GPT-5.5+ completion failures, and fixes Gemini reasoning token tracking.

UPDATE53m ago

Google deprecates temperature, top_p, top_k in Gemini

Google is moving away from traditional developer-tuned LLM hyperparameters—specifically temperature, top_p, and top_k—which will be ignored and rejected by future Gemini models in favor of new control mechanisms.