Moonshot AI27 Jan 2026Model

Kimi K2.5

Made by: Moonshot AI
Released: 27 Jan 2026
Announced in: Kimi K2.5: Visual Agentic Intelligencekimi.ai
Event: Kimi K2.5
API record: /api/v1/models/kimi-k2-5

Benchmark results

Benchmark	Score	Source check
Humanity's Last ExamNo tools · full set, text and image (31.5 text-only); 96K-token completion budget	30.1%	Checked A person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. Corroborated The same number is printed in 1 other lab document: 1
Humanity's Last ExamWith tools · full set, text and image (51.8 text-only); search, code and browsing tools; Hugging Face blocked	50.2%	Checked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
SWE-bench Verified	76.8%	Checked A person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. Corroborated The same number is printed in 5 other lab documents, the first 3 linked: 1 2 3
SWE-bench Proin-house minimal-tool harness; non-thinking mode; average of 5 runs · The source doesn't say whether this is the public or the commercial set	50.7%	Checked A person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. Corroborated The same number is printed in 2 other lab documents: 1 2
GPQA Diamondthinking mode; 96K-token completion budget; average of 8 runs	87.6%	Checked A person checked this number against its source on 20 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.
BrowseCompwith context management, discard-all strategy	74.9%	Checked A person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. Corroborated The same number is printed in 5 other lab documents, the first 3 linked: 1 2 3
no context management: tasks past the context length count as failed	60.6%	Checked A person checked this number against its source on 17 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. Corroborated The same number is printed in 2 other lab documents: 1 2

Sources

Reported by Moonshot AI

Kimi K2.5: Visual Agentic Intelligencekimi.ai · 27 Jan 2026
Our capture of the source

Events

Stories

All models