Kimi K2.6
Moonshot AI · released Apr 14, 2026 · moonshotai/Kimi-K2.6
- Type
- Open weightsCustom licence
- Params
- 1.1T
- Context
- 262K
32B active per word · about 197K words of context · download allowed, licence restricts use
Our take
Written Aug 3, 2026Kimi K2.6 is a large mixture-of-experts model from Moonshot AI that accepts text and images and handles up to 262,144 tokens in a single request. It is the strongest coder in our measured set, with a coding score well above its general chat rating, though its creative writing lags behind both.
Choose this for long-context document analysis or coding tasks where its measured coding score is the strongest signal. It suits high-throughput applications through the fastest host, and budget-conscious inference through the cheapest input offers. Skip it if you need a permissive licence, if creative writing quality matters, or if you cannot tolerate the sixfold throughput gap between providers.
The case for it
- Over one trillion total parameters with only 32 billion active per token, giving an efficient sparsity ratio of roughly 33 to one.
- Highest coding performance in our measured set, with a coding score more than 54 points above its own general chat rating.
- Wide provider choice with input costs varying by a third between the cheapest and most expensive hosts.
- Exceptional throughput available from one provider at 190 tokens per second, more than three times the median across measured offers.
The case against it
- Custom restricted licence, not commercially permissive like Apache or MIT, which limits commercial flexibility.
- Throughput is highly inconsistent across providers, with a 6.6-fold gap between the fastest and slowest measured, and two providers not disclosing throughput at all.
- Creative writing lags its other capabilities, with a creative writing score more than 31 points below its own general chat rating and 85 points below its coding score.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)25th of 143 · 1460.7
CodingWriting and fixing code on its own
Arena Coding19th of 143 · 1514.8
AgenticPlanning, calling tools, staying on task
Arena Agent (IPS)24th of 36 · −0.011
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 K2.6 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 model16 scoresEvery figure we hold, from 16 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 26 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
- $0.80 in / $3.40 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Baidufp4 | $0.59 / $2.48 | 262K | 21 tok/s | No | Yesunknown period | Unknown |
| DigitalOcean Gradient | $0.76 / $3.20 | 262K | 47 tok/s | No | No | Confirmed |
| Novita AI | $0.80 / $3.40 | 262K | not measured | Unknown | Unknown | Unknown |
| ModelRunfp4 | $0.70 / $3.40 | 262K | 111 tok/s | No | No | Unknown |
| SiliconFlowfp8 | $0.77 / $3.40 | 262K | 22 tok/s | No | No | Confirmed |
| Novita AI | $0.80 / $3.40 | 262K | 6 tok/s | No | No | Confirmed |
| Decartfp4 | $0.66 / $3.40 | 262K | 114 tok/s | No | No | Confirmed |
| Inceptronint4 | $0.60 / $3.41 | 262K | 64 tok/s | No | No | Confirmed |
| OpenRouter | $0.60 / $3.41 | 262K | not measured | Unknown | Unknown | Unknown |
| CoreWeavefp4 | $0.65 / $3.41 | 262K | 229 tok/s | No | No | Confirmed |
| Chutesint4 | $0.66 / $3.50 | 262K | 9 tok/s | No | Yesunknown period | Unknown |
| DeepInfrafp4 | $0.75 / $3.50 | 262K | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp4 | $0.75 / $3.50 | 262K | 23 tok/s | No | No | Confirmed |
| Venice AIint4 | $0.75 / $3.50 | 256K | 24 tok/s | No | No | Confirmed |
| Crusoebf16 | $0.70 / $3.50 | 262K | 68 tok/s | No | No | Confirmed |
| Parasailint4 | $0.75 / $3.50 | 262K | 33 tok/s | No | No | Confirmed |
| StreamLakefp8 | $0.85 / $3.60 | 256K | 19 tok/s | No | Yesunknown period | Unknown |
| Basetenfp4 | $0.95 / $4.00 | 262K | 61 tok/s | No | No | Confirmed |
| Moonshot AIint4 | $0.95 / $4.00 | 262K | 26 tok/s | No | No | Confirmed |
| Fireworks AI | $0.95 / $4.00 | 262K | 50 tok/s | No | No | Confirmed |
| AtlasCloudint4 | $0.95 / $4.00 | 262K | 13 tok/s | No | Yesunknown period | Unknown |
| Cloudflare Workers AI | $0.95 / $4.00 | 262K | 31 tok/s | No | Yesunknown period | Unknown |
| Sail Researchint4 | $1.00 / $4.00 | 262K | 15 tok/s | No | No | Confirmed |
| Sail Researchfp8 | $1.00 / $4.00 | 262K | not measured | No | No | Unknown |
| Together AI | $1.20 / $4.50 | 262K | 109 tok/s | No | No | Confirmed |
| Phala | $1.09 / $4.60 | 262K | 32 tok/s | No | No | Confirmed |
Baidu
DigitalOcean Gradient
OpenRouter
DeepInfra
Fireworks AI
Cloudflare Workers AI
Together AIAcross the 26 listings we hold: 23 say they do not train on prompts, 0 say they do and 3 do not say. 16 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 |
|---|---|---|---|
| Baidufp4 | ✓ | ✓ | ✓ |
| DigitalOcean Gradient | ✓ | ✓ | ✓ |
| Novita AI | |||
| ModelRunfp4 | ✓ | ✓ | ✓ |
| SiliconFlowfp8 | ✓ | ✓ | ✓ |
| Novita AI | ✓ | ✓ | ✓ |
| Decartfp4 | ✓ | ✓ | ✓ |
| Inceptronint4 | ✓ | ✓ | ✓ |
| OpenRouter | ✓ | ✓ | ✓ |
| CoreWeavefp4 | ✓ | ✓ | ✓ |
| Chutesint4 | ✓ | ✗ | ✓ |
| DeepInfrafp4 | |||
| DeepInfrafp4 | ✓ | ✓ | ✓ |
| Venice AIint4 | ✓ | ✓ | ✓ |
| Crusoebf16 | ✓ | ✓ | ✓ |
| Parasailint4 | ✓ | ✓ | ✓ |
| StreamLakefp8 | ✓ | ✓ | ✓ |
| Basetenfp4 | ✓ | ✓ | ✓ |
| Moonshot AIint4 | ✓ | ✓ | ✓ |
| Fireworks AI | ✓ | ✓ | ✓ |
| AtlasCloudint4 | ✓ | ✓ | ✓ |
| Cloudflare Workers AI | ✓ | ✓ | ✓ |
| Sail Researchint4 | ✗ | ✗ | ✓ |
| Sail Researchfp8 | |||
| Together AI | ✓ | ✓ | ✓ |
| Phala | ✓ | ✓ | ✓ |
Tool calling: 22 of 26 listings say yes, 1 says no, 3 publish no parameter list. JSON output: 21 of 26 listings say yes, 2 say no, 3 publish no parameter list. Strict schema: 23 of 26 listings say yes, 3 publish no parameter list.
Models people weigh against Kimi K2.6
When we formed this view
Dates behind this page
Prices last checked 13h 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.
- 3 of 26 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.
- 3 of 26 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-K2.6
- Architecture
- Mixture of experts
- Modality record
- text+image->text
- Catalogue slug
- moonshotai-kimi-k2-6