Kimi K2.5
Moonshot AI · released Jan 1, 2026 · moonshotai/Kimi-K2.5
- 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 Sep 30, 2026Kimi K2.5 is a model you can download and run yourself, or reach through one of eight hosted offers. Its custom licence puts conditions on commercial use and redistribution, and its chat preference placings sit in the middle of the field, while its strongest measured signal is on end-to-end coding.
Reach for it on agentic coding work where the model has to resolve real issues in an existing repository, or on long-document work where the request capacity means material need not be split up first. Read the licence before you build a commercial product on it. Skip it if you need a permissive licence, or if you need top-tier chat preference, where it sits in the middle of the field.
The case for it
- 9th of 42 on SWE-bench Verified via mini-SWE-agent as of 19 Feb 2026, which measures the share of real GitHub issues resolved end-to-end inside that harness, so the result is for the model inside that scaffold rather than on its own.
- 262144 tokens of request capacity, so long documents need not be split up first; reliable recall across all of it is unverified in our data.
- OpenRouter, DeepInfra and SiliconFlow all list the same input rate against a higher output rate, so input-heavy work costs less than output-heavy work on those hosts.
The case against it
- The custom licence puts conditions on commercial use and redistribution, so it needs reading before you build on it.
- 71st of 168 on Arena Text (overall) as of 25 Sep 2026 and 77th of 168 on Arena Creative Writing as of 25 Sep 2026, both human preference boards rather than correctness measures.
- 1.1 trillion parameters in total, of which 32 billion are active per token, so this is not a download for a machine of your own.
How good is it?
EverydayGeneral questions and everyday reasoning
Arena Text (overall)71st of 168 · 1430
Also on this board: 1450 (Sep 25, 2026). Read the pair, not the higher one.
CodingWriting and fixing code on its own
Arena Coding47th of 168 · 1503
Also on this board: 1502 (Sep 25, 2026). Read the pair, not the higher one.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent. It is on SWE-bench Verified, in 9th of 42 with 70.8.
WritingDrafting and rewriting prose
Arena Creative Writing77th of 168 · 1390
Arena Creative Writing is the only board that has scored it for this.
Also on this board: 1423 (Sep 25, 2026). Read the pair, not the higher one.
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.
Every published score for this model8 scoresEvery figure we hold, from 8 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?
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.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked between 1 hour and 31 days ago — each listing carries its own date.
Some hosts sell this model at two prices: on their own price list (“direct”) and on their OpenRouter listing (“through OpenRouter”). Where the two differ, the row shows both, each with the date we last read it.
- per 1M tokens
- $0.45 in / $2.25 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $0.45 / $2.25checked 1 hour ago | 262K | not measured | Unknown | Unknown | Unknown |
| DeepInfrafp4Direct | $0.45 / $2.25checked 31 days ago | 262K | not measured | Unknown | Unknown | Unknown |
| SiliconFlowint4Through OpenRouter | $0.45 / $2.25checked 1 hour ago | 262K236K max reply | 22 tok/s | No | No | Confirmed |
| AtlasCloudint4Through OpenRouter | $0.49 / $2.50checked 1 hour ago | 262K236K max reply | 47 tok/s | No | Yesunknown period | Unknown |
| Novita AIDirect and through OpenRouter | $0.60 / $3.00directchecked 1 hour ago$0.57 / $2.85through OpenRouterchecked 1 hour ago | 262K236K max reply through OpenRouter | 41 tok/sthrough OpenRouter | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterNo | DirectUnknownThrough OpenRouterConfirmed |
| Amazon Bedrockus-east-2Through OpenRouter | $0.60 / $3.00checked 1 hour ago | 262K131K max reply | 66 tok/s | No | No | Confirmed |
| Venice AIThrough OpenRouter | $0.53 / $3.33checked 1 hour ago | 256K66K max reply | 16 tok/s | No | No | Confirmed |
DeepInfra
Novita AI
Amazon BedrockAcross the 7 listings we hold: 5 say they do not train on prompts (1 of them only through OpenRouter), 0 say they do and 2 do not say. 4 appear in the zero-retention registry we check (1 of them only through OpenRouter); the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| DeepInfrafp4Direct | |||
| SiliconFlowint4Through OpenRouter | ✓ | ✓ | ✓ |
| AtlasCloudint4Through OpenRouter | ✓ | ✓ | ✓ |
| Novita AIDirect and through OpenRouter | ✓ | ✓ | ✓ |
| Amazon Bedrockus-east-2Through OpenRouter | ✓ | ✗ | ✗ |
| Venice AIThrough OpenRouter | ✓ | ✓ | ✓ |
Tool calling: 6 of 7 listings say yes, 1 publishes no parameter list. JSON output: 5 of 7 listings say yes, 1 says no, 1 publishes no parameter list. Strict schema: 5 of 7 listings say yes, 1 says no, 1 publishes no parameter list.
Models people weigh against Kimi K2.5
When we formed this view
Recent changes
What moved
input +33% ($0.375 → $0.500 per 1M tokens), output +33% ($2.025 → $2.700 per 1M tokens)What moved
input $0.57 → $0.6, output $2.85 → $3 per 1M tokensWhat moved
input $0.57 → $0.6, output $2.85 → $3 per 1M tokensEach date is the day we first saw the change, or the day the maker announced it.
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.
- 1 of 7 listings publishes no parameter list, so what its API accepts is unknown to us.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 2 of 7 listings do not say whether they train on prompts, and 1 answers only through OpenRouter, not for its own listing.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
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-K2.5
- Architecture
- Mixture of experts
- Takes in, gives back
- Text and images in, text out
- Catalogue slug
- moonshotai-kimi-k2-5