Questions? hello@viberation.devGet supportBlogDocsChangelog
Get started

Kimi

Moonshot's open-weight models, built for long documents and agent runs.

Kimi is Moonshot AI's model family, open-weight and aimed squarely at long-horizon work: big context windows, strong tool use, and a Code variant tuned for software tasks. K3 is a multimodal reasoning model with a million-token context. If you have a large repository or a long agent run and do not want to pay frontier prices, this is the family to price against. The Kimi chat app is listed separately under Chats.

At a glance — Kimi K3

Cost
$3.00 in · $15.00 out per 1M tokensMid-range
Memory
1M tokens — about 1,550 pages
Reads
Text · Images · Video
Can do
  • Use tools (agents)
  • Structured output
  • Thinking mode
Good at
Coding: better than 90% · Agentic work: better than 86% · Overall: better than 84%
Reliability
99.95% uptime, last 24h
Released
Jul 16, 2026
Advanced details
API model ID
moonshotai/kimi-k3
Max output
944K tokens
Cached input
$0.30 per 1M
Tokenizer
Other
Coding index
76.2
Agentic work index
50.0
Overall index
43.6

Providers

The same model, hosted by different companies. OpenRouter picks one per request and falls back to the next if it fails.

ProviderInput /1MOutput /1MCached /1MMax outputUptime 24h
Sail Research · fp4$1.20$10.535$0.30944K96.56%
InferenceNet · fp4$1.30$11.40$0.30944K99.76%
Wafer$1.99$15.00$0.30944K99.63%
Makora$2.04$12.75$0.204944K98.88%
Relace · fp4$2.10$10.50$0.21944K99.67%
Morph$2.125$11.90$0.247944K97.82%
Phala$2.55$12.75$0.255944K99.76%
DigitalOcean$2.55$12.95$0.255944K96.45%
DeepInfra · bf16$2.85$14.25$0.28516K98.70%
Chutes · mxfp4$3.00$15.00$0.3066K95.65%
Parasail · fp4$3.00$15.00$0.30944K99.19%
Modal · mxfp4$3.00$15.00$0.30944K99.81%
Together$3.00$15.00$0.30944K99.57%
Fireworks$3.00$15.00$0.30944K99.95%
BaseTen · fp8$3.00$15.00$0.30262K99.51%
Moonshot AI · mxfp4$3.00$15.00$0.30944K99.88%
Fireworks · us$3.30$16.50$0.33944K99.54%
Alibaba$3.45$17.25$0.345944K96.18%
Fireworks · fast$4.50$22.50$0.45944K99.80%

Supported parameters

  • frequency_penalty
  • include_reasoning
  • logit_bias
  • logprobs
  • max_tokens
  • min_p
  • presence_penalty
  • reasoning
  • reasoning_effort
  • repetition_penalty
  • response_format
  • seed
  • stop
  • structured_outputs
  • temperature
  • tool_choice
  • tools
  • top_k
  • top_logprobs
  • top_p

Live from OpenRouter, refreshed hourly. Scores from Artificial Analysis.

Key info

Pricing
Open source
Category
Models
Best for
Intermediate
Made by
Moonshot AI
Weights
Open
Good for
Long documents and long agent runs

More in Models

  • France's frontier lab, with the small models everyone self-hosts.

    Models
  • Meta's current line, built for long multi-agent runs.

    Models
  • The open-weight family everyone else gets benchmarked against.

    Models
  • Google's open models — Gemini's research, small enough to self-host.

    Models