MDL
Kimi K2 Thinking
Kimi
Verified against Artificial Analysis · Jul 13, 2026
Capability
89
percentile index
Price /M
$0.6 / $2.5
in / out
Context
262.1K
tokens
Avg score
67.2
10 benchmarks
AA Intelligence
32.7
index
Speed
171 tok/s
TTFT 0.9s
Released
Nov 6, 2025
About
Kimi K2 Thinking is a open-weights model in the Kimi family from Kimi. Benchmarked on 10 evals, averaging 67.2.
Benchmark scores · 10
avg 67.2 — higher is betterThe research behind this model
all papers mentioning it →Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent▲ 98 on HF · Jun 29, 2026SkillAdaptor: Self-Adapting Skills for LLM Agents from Trajectories▲ 35 on HF · May 31, 2026ComBench: A Benchmark for Rigorous Proof Reasoning and Constructive Realization in Olympiad-Level Combinatorics▲ 19 on HF · Jun 9, 2026HRBench: Benchmarking and Understanding Thinking-Mode Switch Strategies in Hybrid-Reasoning LLMs▲ 15 on HF · May 27, 2026ProCUA-SFT Technical Report▲ 9 on HF · Jun 15, 2026
Mentions matched by name in title/abstract — from the arXiv + HF daily corpus.
Performance & price
Output speed171 tok/s
Latency (TTFT)0.9s
First answer13s
Price in / out$0.6 / $2.5 per 1M
Blended price$1.07 / 1M
Math index94.7
Median across API providers, via Artificial Analysis. Blended = 3:1 input:output. For reasoning models, latency includes thinking time. Methodology →
History
full ledger →Capability over time
Append-only ledger — every observed change to this model's numbers.
API providers
Artificial Analysis$2.5/M