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 Aug 2, 2026Kimi K2.5 is a large mixture-of-experts model from Moonshot AI that handles text and images with up to 262,144 tokens in a single request. It is built for long-document work and coding, with measured coding scores well above its general chat rating.
Choose this for long-context document analysis and coding tasks where its 262,144-token limit and strong measured coding score matter. Use it for image-plus-text workflows that need frontier-scale parameter capacity, or when you want to shop across ten providers for the best rate. Skip it if you need a permissive licence, if creative writing is your main workload, or if you need output pricing under two dollars per million tokens.
The case for it
- Strong measured coding performance: its coding score is 73.3 points above its general text rating on the Arena leaderboard.
- Extremely large request limit for the active parameter count — 262,144 tokens with 32 billion active per forward pass.
- Ten current offers with a 2.4× price spread, giving room to optimise for budget.
- Fastest confirmed option runs at 44 tokens per second, 2.2× the slowest tracked host.
The case against it
- Custom restricted licence, not a permissive standard like Apache or MIT, which limits commercial flexibility.
- Creative writing lags other capability areas by 115.3 points on the measured leaderboard.
- No provider achieves above 44 tokens per second, and two hosts have unverified throughput in our data.
How good is it?
IntelligencePuzzles, maths, exam questions
Arena Text (overall)53rd of 143 · 1431.6
CodingWriting and fixing code on its own
Arena Coding31st of 143 · 1504.7
AgenticPlanning, calling tools, staying on task
Kimi K2.5 is not on Arena Agent (IPS), which is where the rating would come from, so there is no rating here. It is on SWE-bench Verified, in 10th of 39 with 70.8.
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 K2.5 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 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?
- 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 14 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
- $0.57 in / $2.85 out
- Context served
- 262K
- Throughput
- Not measured
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| Chutesint4 | $0.44 / $2.00 | 262K | 20 tok/s | No | Yesunknown period | Unknown |
| DigitalOcean Gradient | $0.38 / $2.02 | 262K | 10 tok/s | No | No | Confirmed |
| DeepInfrafp4 | $0.45 / $2.25 | 262K | 26 tok/s | No | No | Confirmed |
| DeepInfrafp4 | $0.45 / $2.25 | 262K | not measured | Unknown | Unknown | Unknown |
| SiliconFlowint4 | $0.45 / $2.25 | 262K | 55 tok/s | No | No | Confirmed |
| AtlasCloudint4 | $0.49 / $2.50 | 262K | 30 tok/s | No | Yesunknown period | Unknown |
| StreamLakefp8 | $0.54 / $2.70 | 256K | 29 tok/s | No | Yesunknown period | Unknown |
| OpenRouter | $0.57 / $2.85 | 262K | not measured | Unknown | Unknown | Unknown |
| Novita AI | $0.57 / $2.85 | 262K | 26 tok/s | No | No | Confirmed |
| Phala | $0.60 / $3.00 | 262K | 36 tok/s | No | No | Confirmed |
| Novita AI | $0.60 / $3.00 | 262K | not measured | Unknown | Unknown | Unknown |
| Amazon Bedrockus-east-2 | $0.60 / $3.00 | 262K | 65 tok/s | No | No | Confirmed |
| Moonshot AIint4 | $0.60 / $3.00 | 262K | 30 tok/s | No | No | Confirmed |
| Venice AI | $0.56 / $3.50 | 256K | 26 tok/s | No | No | Confirmed |
DigitalOcean Gradient
DeepInfra
Novita AI
Amazon BedrockAcross the 14 listings we hold: 11 say they do not train on prompts, 0 say they do and 3 do not say. 8 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 |
|---|---|---|---|
| Chutesint4 | ✓ | ✓ | ✓ |
| DigitalOcean Gradient | ✓ | ✓ | ✓ |
| DeepInfrafp4 | ✓ | ✓ | ✗ |
| DeepInfrafp4 | |||
| SiliconFlowint4 | ✓ | ✓ | ✓ |
| AtlasCloudint4 | ✓ | ✓ | ✓ |
| StreamLakefp8 | ✓ | ✓ | ✓ |
| OpenRouter | ✓ | ✓ | ✓ |
| Novita AI | ✓ | ✓ | ✓ |
| Phala | ✓ | ✓ | ✓ |
| Novita AI | |||
| Amazon Bedrockus-east-2 | ✓ | ✗ | ✗ |
| Moonshot AIint4 | ✓ | ✓ | ✓ |
| Venice AI | ✓ | ✓ | ✓ |
Tool calling: 12 of 14 listings say yes, 2 publish no parameter list. JSON output: 11 of 14 listings say yes, 1 says no, 2 publish no parameter list. Strict schema: 10 of 14 listings say yes, 2 say no, 2 publish no parameter list.
Models people weigh against Kimi K2.5
When we formed this view
Dates behind this page
Prices last checked 14h 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 14 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 14 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-K2.5
- Architecture
- Mixture of experts
- Modality record
- text+image->text
- Catalogue slug
- moonshotai-kimi-k2-5