LLM Open Weight Modified MIT

Kimi K2 Thinking

by Moonshot AI

Prices last updated November 22, 2026

Kimi K2 Thinking, an open-weight Moonshot AI model available via OpenRouter.

Parameters
1T MoE (32B active)
Context
262k
Providers
4
Cheapest Input
$0.470
Function Calling Streaming Chat / InstructReasoning
HuggingFace

Price Comparison

Compare pricing across 4 providers serving Kimi K2 Thinking

Provider Input $/1M Output $/1M Context Speed TTFT Status
DeepInfra
DeepInfra
$0.470 ● $2.00 131k — — Live
OpenRouter
OpenRouter
$0.600 $2.50 262k — — Live
Novita AI
Novita AI
$0.600 $2.50 262k — — Live
Infron
Infron
$0.600 $2.50 262k — — Live
● Cheapest ● Fastest Prices in USD per 1M tokens

Benchmarks

Independent results for Kimi K2 Thinking, as each source publishes them. Where a source tested several versions (effort levels, thinking modes, dated snapshots), the best one is shown and named under "As tested". Scores from different sources use different scales and aren't comparable with each other.

BenchmarkScoreRankAs tested
Overall
One score combining a model's results across many benchmarks, on a common scale.
145.5±1.8
#95 of 258
Kimi K2 Thinking
Reasoning
GPQA DiamondEpoch AI
Graduate-level multiple-choice questions in biology, physics and chemistry, written to be hard to look up.
84.2%±4.1
#86 of 232
kimi-k2-thinking-turbo
Nov 2025
Math
Competition maths problems from the OTIS mock AIME exams (2024-2025).
83.1%±9.4
#96 of 215
kimi-k2-thinking-turbo
Nov 2025
Coding
Text Arena votes on coding prompts only.
1,486±5
15,131 votes
#110 of 421
Kimi K2 Thinking Turbo
Nov 2026
Human votes on web apps the models build from a prompt.
1,322±8
12,997 votes
#139 of 161
Kimi K2 Thinking Turbo
Nov 2026
Chat
Text ArenaLMArena
Human votes on chat answers across all kinds of prompts (style-controlled).
1,430±3
62,866 votes
#123 of 426
Kimi K2 Thinking Turbo
Nov 2026

Epoch AI: Epoch runs each benchmark itself on the models it evaluates, with the same harness for every model. The Epoch Capabilities Index combines many benchmarks into one score. Data: Epoch AI, 'Capabilities & benchmarking', epoch.ai/benchmarks (CC BY 4.0).

LMArena: People compare two anonymous models side by side and vote for the better answer; ratings come from those votes (Bradley-Terry, style-controlled for text and vision). Data: LMArena (arena.ai) leaderboard dataset (CC BY 4.0).

Rank is among all models the source ranks, including ones we don't list.

Subscribe to our newsletter

News and updates, 1–2 emails per month. No spam.