NEWMEN
← back to all models

MoonshotAI: Kimi K2 Thinking

moonshotai/kimi-k2-thinking

Moonshotopen-weight262K context3 providersIntelligence 87.9

Cheapest provider

$0.60 / 1M

AtlasCloud

Fastest provider (p95)

No throughput data yet — populated as traffic accumulates

Intelligence (composite)

87.9

MMLU-Pro · HumanEval · math · GPQA

Per-provider performance

Latency / throughput / uptime measured across providers over the last 30 minutes of live traffic. Atlas’s router weighs these per call (with the eval-gate signal) when picking a variant for Standard and Batch tiers.

ProviderQuantInput $/1MOutput $/1MLatency p50 / p95Throughput p50 / p95Uptime 30mSuccess
AtlasCloudq4· int4$0.6000$2.5000
Googleundisclosed$0.6000$2.5000100.00%
Novitafull· bf16$0.6000$2.500092.86%

“—” means live telemetry hasn’t accumulated enough recent traffic for that endpoint. “undisclosed” means the provider serves the model but doesn’t expose the quantization label (typically running fp8 / int8 internally).

Intelligence breakdown

Composite score is a weighted average of public benchmarks (30% MMLU-Pro, 25% code pass@1, 25% math, 20% GPQA). Numbers come from model cards and the Artificial Analysis intelligence harness; missing components are renormalised over what’s present.

MMLU-Pro

84.6

broad reasoning

Code

pass@1 (HumanEval / LiveCodeBench)

AIME 2025

94.5

math accuracy

GPQA Diamond

84.5

hard reasoning

Source: Moonshot AI — Kimi K2 Thinking model card (MMLU-Pro 84.6, GPQA-Diamond 84.5, AIME25 94.5 no-tools); huggingface.co/moonshotai/Kimi-K2-Thinking

How Atlas routes MoonshotAI: Kimi K2 Thinking

  • Realtime tier — direct passthrough at the upstream’s native precision. Best for hard latency / quality guarantees.
  • Standard tier — Atlas picks the cheapest provider variant whose quantization has stayed green on your operation’s eval gates. For MoonshotAI: Kimi K2 Thinking that’s currently AtlasCloud at $0.60/1M.
  • Batch tier — async, biggest discount. Roadmapped to use provider batch APIs (OpenAI / Anthropic) where available and queued spot capacity for open-weight workloads.