Model
Model explorer

Kimi K3.1 Pro

CLOSED
Moonshot AI · Kimi K3 family · released Oct 23, 2026

First model of Moonshot's K3.1 series, launched Oct 23, 2026 on Kimi.ai, Kimi Work, Kimi Code and the Kimi API, alongside a 2-3x faster Kimi K3.1 Pro Highspeed. Moonshot calls it a post-trained Kimi K3 with much more RL compute and says full weights will ship by Nov 4, 2026, so it is filed as closed until they do (MiniMax M3.1 Flash precedent). Moonshot also says it removed cyber-related data from mid-training and RL for the open-weight release. API price $2/$10 per Mtok (Artificial Analysis, from Kimi's API; Kimi's own pricing table renders client-side and could not be read). Scores are Moonshot's at max effort. NOT FILED: OfficeQA Pro v2 (64.4; the corpus officeqa-pro column holds v1 figures, e.g. Opus 5.5 67.7 vs 62.2 on v2), Frontier Finance (58.6), ProgramBench (binary-score variant), GDPval-AA v2.1 (AA re-based scale), Agents' Last Exam, GDP.pdf, ChartoGraphy, ZeroBench, HLE without tools, and Kimi's internal suites (Kimi Code Bench, Hallu Bench, xdev-bench-hard); DeepSWE v1.1 (75.5) and MMMU-Pro (84.1) appear in only one transcription of the image grid. Kimi K3.1 Pro Highspeed is not filed separately: no benchmark table and no first-party price could be read.

ReasoningCodingVisionFunction callingTool useAgentic
3169.9
Elo · rank #1
Parameters
2800B
Active params
Undisclosed
Context
1.024M tokens
Architecture
Post-trained version of Kimi K3 (same 2.8T-parameter MoE base) with substantially larger multi-domain joint RL; 1M-token context; always-on thinking with reasoning_effort low, high or max (default max); text, image and video input
License
Proprietary API for now (Moonshot says full model weights will be released by Nov 4, 2026)
Languages
—
API price (in/out)
$2 / $10
Modalities
text · vision · video
Benchmark results
Bar shows position within the tracked field; marker = field best
best: GPT-6 Astra · 95.9%
BrowseCompAgents92.1%#1
best: this model · 92.1%
GPQA DiamondReasoning94.2%#6
best: GPT-6 Astra · 96.0%
Humanity's Last ExamReasoning70.0%#1
best: this model · 70.0%
OSWorld 2.0Agents79.6%#2
best: Claude Opus 5.5 · 81.8%
SWE-MarathonCoding51.7%#3
best: Muse Wonder Full · 57.5%
Terminal-Bench 4.0Agents63.6%#3
best: Claude Sonnet 5.5 · 70.6%
best: GPT-6 Astra · 62.9%
Run it locally
Closed weights — available via API only. No local deployment.
Input / M tok
$2
Output / M tok
$10
API price $2/$10 · each benchmark row carries its own source badge (see methodology)