K
Kimi K3
The largest open-weight model, with GPQA 93.1 closing in on the closed leaders
Intelligence v5
— v4.3.2 43.6
Leaderboard rank
—
Input / output
$2.00/$15.00
Blended 3:1
$5.25
Output speed
30 tok/s
Value
—
Capability profile Normalised to the field maximum
Specifications
Vendor
Moonshot AI
Released
2026-07-27
Weights
Open (Kimi K3 License)
Context
1,000,000 tokens
Max output
Undisclosed
Parameters
2780B total / 104B active
Input price
$2.00 / 1M
Output price
$15.00 / 1M
Cache hit
n/a
Output speed
30 tok/s
Intelligence
43.6（v4.3.2）
Benchmark scores
How it is measured →Bars are scaled to the field maximum for that benchmark.
| Benchmarks | Score | Vs. best in field | Gap |
|---|---|---|---|
| AA Intelligence Index v5 | — | — | not indexed |
| AA Intelligence Index v4.3.2 Intelligence index (legacy) | 43.6 | 76% | −14.0 pts |
| GPQA Diamond Graduate-level science reasoning | 93.1 | 97% | −3.3 pts |
| Terminal-Bench 4.0 Terminal-agent tasks | 17.2 | 26% | −48.0 pts |
| LiveBench Design Document generation (DOCX/PPT/PDF) | 77.2 | 89% | −9.7 pts |
| Vals Index | — | — | not indexed |
Assessment
Open-weighted on 2026-07-27: Hugging Face lists 2.78T parameters with 104B active, native text/image/video input and 1M context. GPQA Diamond 93.1% is within 3 points of the best closed result. The licence requires a visible Kimi K3 attribution once you pass 100M monthly users or $20M monthly revenue.
At the same price
The four models closest on blended price, side by side.
| Model | v5 | Blended | tok/s | Score gap |
|---|---|---|---|---|
| Claude Sonnet 5.5Anthropic | 54.2 | $4.00 | 135 | — |
| GPT-6.1 SolOpenAI | 54.5 | $4.00 | 77 | — |
| Gemini 4 CarbonGoogle | 59.4 | $4.00 | 87 | — |
| Gemini 4 ArgonGoogle | 50.4 | $4.00 | 80 | — |
Compare side by side
Line it up against other models on price, speed and every benchmark.
The comparison tool lines up to 4 models side by side with switchable normalisation — built for purchasing decisions.
Open the comparison tool → Back to leaderboard