Kimi K3
Moonshot AI · released Jun 13, 2026 · moonshotai/Kimi-K3
- Type
- Open weightsCustom licence
- Params
- 2.8T
- Context
- 1M
104B active per word · about 786K words of context · download allowed, licence restricts use
Our take
Written Aug 2, 2026Kimi K3 is Moonshot AI's flagship coding model with nearly 2.8 trillion parameters and a one-million-token request limit. Its measured coding performance sits at the frontier, though its custom licence carries restrictions and its agentic scores are thinly measured.
Choose this for frontier-level coding work where the WebDev benchmark score matters, or for long-context workflows that need the full one-million-token window. Pick it when throughput is critical and you can route to the fastest host. Skip it if you need a permissive licence for commercial redistribution, if you rely on agentic task performance, or if you need verified vision quality beyond basic image acceptance.
The case for it
- Massive scale with selective activation: 2,779.9 billion total parameters, 104 billion active per token.
- Strong measured coding performance, with five WebDev runs clustering tightly around 1681–1682 Elo.
- One-million-token request limit, matching the largest proprietary windows.
- Throughput varies 2.68× across hosts at the same price, so routing choice matters.
The case against it
- Custom open-restricted licence, not Apache 2.0 or MIT — check terms before commercial use or redistribution.
- Agentic benchmark is thinly measured and low: only two runs, peaking at 0.0991.
- Accepts images but no vision benchmark scores are held.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)7th of 143 · 1485.3via Max
CodingWriting and fixing code on its own
Arena Coding5th of 143 · 1531.5via Max
AgenticPlanning, calling tools, staying on task
Arena Agent (IPS)4th of 36 · 0.099via Max
WritingWe do not rate this
Two boards come close and neither tests writing: Arena Creative Writing asks people which of two replies they prefer, and LiveBench Language tests whether a model understood a passage. So we show where Kimi K3 placed and give it no mark out of five.
These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.
Every published score for this model15 scoresEvery figure we hold, from 15 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
- Fits in memory
- weights load entirely on the card
- Spills to system RAM
- some weights offload; much slower
- Too large
- will not load even with offload
- est
- size is calculated; the verdict could change by 10%
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Radeon RX 7900 XT · 20 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Memory use by level
Against a 24 GB card.
All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.
Check against your own machine → · All 71 devices, with every size →
Or rent it from someone else
Cheapest of 17 live listings. Picked at the widest standard context we hold, within one quantisation slice, so the numbers beside it are a price one host actually charges.
- per 1M tokens
- $3.00 in / $15.00 out
- Context served
- 1M
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Morph | $2.90 / $14.00 | 1M | 46 tok/s | No | No | Confirmed |
| Novita AI | $3.00 / $15.00 | 1M | not measured | Unknown | Unknown | Unknown |
| Modalmxfp4 | $3.00 / $15.00 | 1M | 51 tok/s | No | No | Confirmed |
| Fireworks AI | $3.00 / $15.00 | 1M | 37 tok/s | No | No | Confirmed |
| Modalfp4 | $3.00 / $15.00 | 1M | not measured | No | No | Unknown |
| OpenRouter | $3.00 / $15.00 | 1M | not measured | Unknown | Unknown | Unknown |
| Together AI | $3.00 / $15.00 | 1M | 35 tok/s | No | No | Confirmed |
| DigitalOcean Gradient | $3.00 / $15.00 | 1M | 17 tok/s | No | No | Confirmed |
| Chutesmxfp4 | $3.00 / $15.00 | 1M | 20 tok/s | No | Yesunknown period | Unknown |
| Basetenfp8 | $3.00 / $15.00 | 1M | 23 tok/s | No | No | Confirmed |
| Moonshot AIint4 | $3.00 / $15.00 | 1M | not measured | No | No | Unknown |
| Moonshot AImxfp4 | $3.00 / $15.00 | 1M | 23 tok/s | No | No | Confirmed |
| Waferfp8 | $3.00 / $15.00 | 912K | 40 tok/s | No | No | Unknown |
| Nebius AI Studiofp4 | $3.00 / $15.00 | 1M | 44 tok/s | No | No | Unknown |
| Morphfast tier | $6.00 / $22.50 | 1M | 39 tok/s | No | No | Confirmed |
| Fireworks AIfast tier | $4.50 / $22.50 | 1M | 73 tok/s | No | No | Confirmed |
| Waferfast tierfp8 | $4.50 / $22.50 | 1M | 55 tok/s | No | No | Unknown |
Morph
Fireworks AI
OpenRouter
Together AI
DigitalOcean Gradient
Nebius AI StudioAcross the 17 listings we hold: 15 say they do not train on prompts, 0 say they do and 2 do not say. 9 appear in the zero-retention registry we check; the rest are unknown to us rather than confirmed either way.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
- ✓
- Supported
- ✗
- Not supported
- Not published
- host gave no parameter list
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| Morph | ✓ | ✓ | ✓ |
| Novita AI | |||
| Modalmxfp4 | ✓ | ✓ | ✓ |
| Fireworks AI | ✓ | ✓ | ✓ |
| Modalfp4 | ✓ | ✓ | ✓ |
| OpenRouter | ✓ | ✓ | ✓ |
| Together AI | ✓ | ✓ | ✓ |
| DigitalOcean Gradient | ✓ | ✓ | ✗ |
| Chutesmxfp4 | ✗ | ✓ | ✓ |
| Basetenfp8 | ✓ | ✗ | ✗ |
| Moonshot AIint4 | |||
| Moonshot AImxfp4 | ✓ | ✓ | ✓ |
| Waferfp8 | ✗ | ✓ | ✓ |
| Nebius AI Studiofp4 | ✓ | ✗ | ✗ |
| Morphfast | ✓ | ✓ | ✓ |
| Fireworks AIfast | ✗ | ✓ | ✓ |
| Waferfast · fp8 | ✓ | ✓ | ✓ |
Tool calling: 12 of 17 listings say yes, 3 say no, 2 publish no parameter list. JSON output: 13 of 17 listings say yes, 2 say no, 2 publish no parameter list. Strict schema: 12 of 17 listings say yes, 3 say no, 2 publish no parameter list.
Models people weigh against Kimi K3
When we formed this view
Dates behind this page
Prices last checked 34h ago
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- 2 of 17 listings publish no parameter list, so what their API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 2 of 17 listings do not say whether they train on prompts.
Licence and identifiers
What the licence allowsCustom licence, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Custom licence
This model ships custom license terms that don't map to a known template. We haven't parsed them, so commercial use, redistribution and derivatives are unverified — review the original terms before shipping.
Identifiers
- Hugging Face
- moonshotai/Kimi-K3
- Architecture
- Mixture of experts
- Modality record
- text+image->text
- Catalogue slug
- moonshotai-kimi-k3