NaiveAI ships open-weights Naive-N0.5-Flash for fast coding
Original title: NaiveAI Promises 2,100 Tokens per Second. How Well Does It Code?
A small Latent Space test saw high server-side throughput and very long hard-question traces, without grading accuracy.
NaiveAI, a Chinese company, released Naive-N0.5-Flash, an open-weights coding model continued from Xiaomi’s MiMo-V2.5 and served with NaiveRT. The vendor pitches about 2,100 tokens per second and recursive self-improvement. Latent Space compared ultrafast mode with DeepSeek V4.1 Flash 1017 and GLM 5.3 Flash on 30 questions and did not score correctness. Published cards put Naive ahead on NL2Repo-Bench and between the two on other agent benchmarks.
Unconfirmed: Naive throughput is server-side while rivals were measured via OpenRouter, so speed multiples are not claimed.
- MiMo-V2.5 is a 310B sparse MoE model with 15B active parameters.
- NaiveAI cites about 3.25T extra tokens, mostly at a 1M context length.
- NL2Repo-Bench scores: Naive-N0.5-Flash 71.9, DeepSeek 68.2, GLM 56.3.
- Median thinking response was 3.1 seconds; means were pulled up by long traces.
From the original article
9 min read at latent.space
NaiveAI could potentially top Artificial Analysis’s speed rankings. Our hands-on test suggests its coding quality holds up against DeepSeek.
By Gabriele Beltramo · latent.space
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.