benchmarkIMAGE

Flint v1.0 在 WikiBench 上排第 7，领先 MiniMax M3.1 Flash

Recursive SI 的 Flint v1.0 在 WikiBench 14 个模型中排第 7，均分 4.56，工具调用仅 0.91，但仍排在 MiniMax M3.1 Flash 之前。

Wayne Lowry

@wikiwayne6 天前收录原推发布于 2026.10.22

Flint v1.0 在 WikiBench 上排第 7，领先 MiniMax M3.1 Flash（第 1 张，共 2 张）

Flint v1.0 在 WikiBench 上排第 7，领先 MiniMax M3.1 Flash（第 2 张，共 2 张）

Wayne Lowry @wikiwayne原推发布于 2026.10.22

Flint v1.0 from Recursive SI just ran WikiBench: 7th of 14, average 4.56. Free on OpenRouter. It's sold as a coding agent, yet tool use scored 0.91. Where it held up: • Humanized writing 8.25 • Research 7.60 • Hallucination resistance 7.50 • Security 3.75, tied for best in class It struggled on tool use (0.91) and computation (2.00). It still finished ahead of Laguna S 2.1, MiniMax M3.1 Flash, and North Mini Code. Would you run a free model that writes well but can't call tools?

x.com/wikiwayne/status/2113222390628929676采集于 2026.10.23 12:10