kimi-k3Moonshot AI
- endpoints
- 5
- regions
- 2
- best tok/s
- 92
- api
- chat
- released
- July 2026
- eu routing
- available
capabilities 5/8
| # | flags | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| 1 |  Fireworks AI | Global | 1M | $3.00 | $15.00 | $0.30 | 2.01s | 54.0 | zdr | |
| 2 |  Moonshot AI | Global | 1.0M | $3.00 | $15.00 | $0.30 | 5.86s | 38.0 | - | zdr |
| 3 |  Nebius AI | EU | 1.0M | $3.00 | $15.00 | - | 1.19s | 43.0 | zdr | |
| 4 |  sference | EU | 1.0M | $3.00 | $15.00 | $0.45 | 75ms | 92.0 | zdr | |
| 5 |  TensorX Ltd. | EU | 1.0M | $3.00 | $15.00 | $0.75 | 1.30s | 38.0 | zdr |
| provider | now | low | high | change |
|---|---|---|---|---|
| sference | 75ms | 30ms | 506ms | -82% |
| Nebius AI | 1.19s | 556ms | 32.95s | -67% |
| TensorX Ltd. | 1.30s | 990ms | 2.11s | -13% |
| Fireworks AI | 2.01s | 1.09s | 1.54s | +8.6% |
| Moonshot AI | 5.86s | 2.00s | 6.42s | +179% |
median wait before the first token lands. this is the number a user feels, and it moves with provider load through the day.
Lines are steps: each sample is one week of traffic held flat, not a slide from the week before. Click a name in the key to drop it from the plot. Measured through Sep 12.
| eval | Score as a share of 100 | /100 |
|---|---|---|
| GPQA Diamond | 93.5 | |
| Coding Index | 76.2 | |
| SciCode | 59.5 | |
| Humanity's Last Exam | 46.9 | |
| Intelligence Index | 43.8 | |
| 5 evals | mean | 64.0 |
Scores from Artificial Analysis and public leaderboards, normalised to 100. Bar colour is the band, not the rank: green 80 and up, blue 55 and up, amber 30 and up. Benchmarks measure narrow skills, so test on your own workload before committing. Released 2026-07-16.
kimi-k3requesty routes this id across every provider serving kimi-k3, picking on price and health and failing over automatically. the id stays valid when a provider changes underneath it
Base url is https://router.requesty.ai/v1 for every id here. One key reaches the whole catalog.
from openai import OpenAI client = OpenAI( api_key="YOUR_REQUESTY_API_KEY", base_url="https://router.requesty.ai/v1",) response = client.chat.completions.create( model="kimi-k3", messages=[ {"role": "user", "content": "Explain quantum computing in one paragraph."}, ],) print(response.choices[0].message.content)
Change the base url, use your Requesty key, set the model to any id in the call pane. Existing OpenAI SDK code needs no other edit, and the same key reaches every other model in the catalog, so switching later is a one-parameter change. Browse all models
Kimi’s most capable flagship model to date, with 2.8 trillion parameters. It is built on Kimi Delta Attention (KDA), a hybrid linear attention mechanism, and Attention Residuals, with native visual understanding and a 1M-token context window. It is the world’s first open-source model in the 3-trillion-parameter class, designed for frontier intelligence scenarios including long-horizon coding, knowledge work, and reasoning.
questions 7
Which providers serve kimi-k3?
How much does kimi-k3 cost?
Is kimi-k3 open weights?
What is the context window of kimi-k3?
Which provider is fastest for kimi-k3?
How do I use kimi-k3 with the OpenAI SDK?
Should I call kimi-k3 by its managed id or a provider id?
more from moonshot ai 5
| model | context | from /M | providers |
|---|---|---|---|
| kimi-k2.7-code | 262K | $0.75 | 4 |
| kimi-k2.6 | 262K | $0.60 | 5 |
| kimi-k2.5 | 262K | $0.45 | 3 |
| kimi-k2-instruct | 131K | $0.57 | 1 |
| kimi-k2 | 262K | $0.60 | 1 |
route kimi-k3 through one endpoint
One key for 5 providers on this model and 600+ others. No markup on provider prices, automatic failover, caching built in. Weekly aggregates in the measured pane come from production traffic routed through Requesty, one line per provider on a shared axis. Methodology