Cohere x Palantir model benchmark mocked online: 670 tokens/s but questionable output
XiRw · reddit · 2026-10-31
A Reddit user criticized the Cohere and Palantir partnership's delivered results as "embarrassing," sharing a benchmark screenshot: the north-mini-code model generating a pelican-on-a-bicycle SVG at up to 670.50 tokens/s using Q8 quantization and F16 KV Cache. The post sparks discussion about the quality-versus-speed tradeoff in the partnership's deployment.
More from Models
- OpenAI's $100 Pro looks like a steal vs Anthropic's $200 Ultra: doubled quotas, no 5-hour window — gethackteam · 2026-10-31
- ChatGPT desktop voice mode still slow and unreliable, hands-on test finds — koltregaskes · 2026-10-31
- Google fixes Antigravity IDE to add Argon; early user calls it a good model — zacharynado · 2026-10-31
- Mistral open-sources Le Chonk, tops FrontierCode among open models — _AndrewZhao · 2026-10-31
- Blogger predicts OpenAI major release as Anthropic preps Fable 5.5 and new Opus — kimmonismus · 2026-10-31
- DeepMind: RL on simulated patients cuts AI psychiatry missed risk escalation from 30.5% to 12.9% — rohanpaul_ai · 2026-10-31