← Back to directory
Moonshot AI
Kimi K2
Released Jul 11, 2025 · knowledge cutoff 2025-04
- Status
- Superseded
- Location
- China
- Type
- LLM
- Modality
- Text
- Context window
- 131K
- Max output
- 16K
- Speed
- —
- Price ($/MTok in / out)
- $0.60 / $2.50
- Cost per task
- —
- Open weights
- Yes
Benchmarks7 of 10 reported
MMLU-Proretired
81.1%
GPQA Diamond
75.1%
SWE-bench Verified
65.8%
SWE-bench Pro
27.67%
AIME (latest)retired
49.5%
Humanity's Last Exam
4.7%
LMArena Elo
1385
Terminal-Bench 2.1
—
GDPval-AA v2
—
ARC-AGI-2
—
Strengths & weaknesses
Strengths
- Built to call tools rather than to chat — clean calls, sensible arguments, few loops
- A distinct writing voice that avoids the flat assistant register
- Answers immediately, with no reasoning wait
Weaknesses
- No deliberation at all, so hard maths and multi-step logic fail outright
- Large enough that self-hosting is a serious infrastructure commitment
Licence is Moonshot's Modified MIT: MIT plus a 'Kimi K2' UI-attribution clause above 100M MAU or US$20M monthly revenue, classified permissive. SWE-bench Pro 27.67% is Scale's public-set leaderboard figure for kimi-k2-instruct — third-party, not Moonshot-reported.
For developers
API model strings
Not researched
Licence
Modified MIT License · Permissive · commercial use permitted
Retirement
No retirement announced
Lineage
No recorded predecessor