Kimi K3
MoonshotReleased Jul 16, 2026Open weights · Modified MITkimi-k3
- Intelligence (Epoch)
- 158 · 15th of 111
- Reasoning
- yes, with effort levels
- Knowledge cutoff
- April 2026
- Context
- 1M tokens
- Max output
- 1M tokens
- Input
- text, image
- API features
- tools, structured outputs, JSON mode, seed, logprobs
- Moonshot's API
- yes, as kimi-k3
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from Moonshot · cutoff: aiknowledgecutoff.com
Intelligence for the price
1 model scores higher for the same price or less; it ranks 2 of the 93 models at its price or below.
Epoch Capabilities Index against the lowest input + output price per M, log scale · prices Oct 23. Score: Epoch AI, CC BY 4.0, 158 with an interval of 155 to 161.
Against its rivals
Its previous version, the best of its price range and the top score
- Kimi K3this model$11.35158
- Claude Opus 5.5top score$24.00169
- GPT-6.1 Solsame price range$12.00168
- Claude Sonnet 5.5same price range$12.00166
- Kimi K2.6previous version$2.40151
Same level for 60× less: GLM-5.3-Flash at StreamLake, $0.19 per M input + output · their Epoch intervals overlap (153 against 158).
Bars: Epoch Capabilities Index (Epoch AI, CC BY 4.0), drawn from 100, below the lowest score in the catalogue · price: lowest input + output per M.
Speed by provider
Output speed of Kimi K3 at each host, fastest first (the 10 fastest of 15)
- Amazon Bedrock73 tok/s3.10 s to first token
- Relace70 tok/s1.61 s to first token
- Modal69 tok/s1.10 s to first token
- InferenceNet67 tok/s0.60 s to first token
- Parasail62 tok/s0.84 s to first token
- Morph59 tok/s1.17 s to first token
- AkashML53 tok/s1.02 s to first token
- DeepInfra42 tok/s1.31 s to first token
- BaseTen42 tok/s2.57 s to first token
- Together AI41 tok/s1.00 s to first token
Speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · a snapshot of real traffic before the daily read, not a daily average · bars from zero.
Run it yourself
open weights, Modified MIT · too large for the 6 workstations in our guideSelf-host guide ›- Parameters
- 2780B
- 104B active per token (MoE)
- Memory needed
- 1668 GB
- in INT4 · 3336 GB in INT8
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 1668 GB | multi-node, over 640 GB | smallest footprint, minor quality loss |
| INT8 (Q8) | 3336 GB | multi-node, over 640 GB | near-lossless |
| FP16 (full) | 6672 GB | multi-node, over 640 GB | reference quality |
At home: 1668 GB in INT4 is beyond any desktop or Mac in our machines table, so it means a multi-GPU server.
Memory ≈ parameters × bytes per parameter × 1.2 · cloud price: on-demand rate, running 24/7 · builds: GGUF, AWQ, MLX, GPTQ, Ollama · break-even calculator
| Provider | Input | Output | Cache ↑ | tok/s | First token | Uptime 24h | Checked |
|---|---|---|---|---|---|---|---|
| Chutesmxfp4 · 1M ctx · 66K out | $1.85 | $9.50 | $0.18 | 19 | 3.67 s | 100% | Oct 23 |
| Phalaprecision not read: host terms · 1M ctx−15% | $2.55 | $12.75 | $0.26 | 22 | 6.65 s | 99.9% | Oct 23 |
| DeepInframxfp4 · 1M ctx | $2.85 | $14.25 | $0.28 | 42 | 1.31 s | 99.9% | Oct 23 |
| Relacefp4 · 1M ctxreprices often | $2.40 | $14.00 | $0.30 | 70 | 1.61 s | 99.9% | Oct 23 |
| DigitalOcean1M ctx | $3.00 | $15.00 | $0.30 | 32 | 1.47 s | 100% | Oct 23 |
| Parasailfp4 · 1M ctx | $3.00 | $15.00 | $0.30 | 62 | 0.84 s | 98.8% | Oct 23 |
| Modalmxfp4 · 1M ctx | $3.00 | $15.00 | $0.30 | 69 | 1.10 s | 100% | Oct 23 |
| Together AIprecision not stated · 1M ctx | $3.00 | $15.00 | $0.30 | 41 | 1.00 s | 99.9% | Oct 23 |
| Fireworks AI1M ctx | $3.00 | $15.00 | $0.30 | — | — | — | Oct 20 |
| BaseTenfp8 · 1M ctx · 256K out | $3.00 | $15.00 | $0.30 | 42 | 2.57 s | 99.6% | Oct 23 |
| Moonshot AImxfp4 · 1M ctx | $3.00 | $15.00 | $0.30 | 27 | 6.17 s | 100% | Oct 23 |
| Novita AIprecision not stated · 1M ctx · 16K out | $3.00 | $15.00 | $0.30 | — | — | — | Sep 5 |
| InferenceNetfp4 · 1M ctxreprices often | $3.00 | $15.00 | $0.30 | 67 | 0.60 s | 100% | Oct 23 |
| Amazon Bedrockprecision not read: host terms · 1M ctx · 125K out | $3.00 | $15.00 | $0.30 | 73 | 3.10 s | 99.3% | Oct 23 |
| Alibabaprecision not read: host terms · 1M ctx · 128K out | $3.45 | $17.25 | $0.34 | 30 | 1.67 s | 99.8% | Oct 23 |
| Morphfp8 · 1M ctxreprices often | $1.25 | $14.90 | $0.60 | 59 | 1.17 s | 99.9% | Oct 23 |
| AkashMLfp4 · 1M ctx | $1.20 | $14.00 | $1.20 | 53 | 1.02 s | 100% | Oct 23 |
10 of 17 providers · prices per million tokens, read Oct 23, 06:00 UTC · speed: OpenRouter, median of the last 30 min, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · −N%: a discount off the host's list price · under each host: weights precision, context and max output as served, the maker's value in brackets when the host's differs
| Provider | Weights | Context | Max output | Tools | Structured outputs | Reasoning |
|---|---|---|---|---|---|---|
| Chutes | mxfp4 | 1M | 66K | no | yes | yes |
| AkashML | fp4 | 1M | not stated by the host | yes | yes | yes |
| Phala | not read: host terms | 1M | not read: host terms | yes | yes | yes |
| Morph | fp8 | 1M | not stated by the host | yes | yes | yes |
| Relace | fp4 | 1M | not stated by the host | yes | no | yes |
| DeepInfra | mxfp4 | 1M | not stated by the host | yes | yes | yes |
| DigitalOcean | not checked yet | 1M | not checked yet | yes | yes | yes |
| Parasail | fp4 | 1M | not read: host terms | yes | yes | yes |
| Modal | mxfp4 | 1M | not stated by the host | yes | yes | yes |
| Together AI | not stated by the host | 1M | not stated by the host | yes | yes | yes |
| Fireworks AI | not checked yet | 1M | not checked yet | — | — | — |
| BaseTen | fp8 | 1M | 256K | yes | no | yes |
| Moonshot AI | mxfp4 | 1M | not checked yet | yes | yes | yes |
| Novita AI | not stated by the host | 1M | 16K | — | — | — |
| InferenceNet | fp4 | 1M | not stated by the host | yes | yes | yes |
| Amazon Bedrock | not read: host terms | 1M | 125K | yes | no | yes |
| Alibaba | not read: host terms | 1M | 128K | yes | yes | yes |
Weights, context and max output: read at each host's own catalogue or docs where its terms allow, otherwise as the host declares to OpenRouter; hover a cell for its source and date · tools, structured outputs, reasoning: as declared to OpenRouter · not verified by us
Kimi K3 spec sheet, field by field, with sources
Kimi K3 spec sheetmaker-declared
What Moonshot declares about the model itself — read from its own weights repository and documentation, not from any host. What each endpoint actually serves is in the pricing above: the two can differ, and both are true.
- Context window
- 1,048,576 tokens
- Parameters
- 2,779,931,837,184≈ 2.78T
- Active experts
- 16 of 896 active per token
- Weight precision
- not published by the maker
- License
- Modified MIT — open weights, conditionalcuratedScope: the licence on the published weights, and nothing else. It is not the maker's acceptable-use policy, which is a separate document, and it is not the contract you sign with whichever host you call — hosts set their own terms, which this index does not read.
- Knowledge cutoff
- not published by the maker≈ April 2026 third-party estimate — not the maker's figurecurated
- Modalities
- text · image
- Max output (official)
- 1,048,576 tokensmaker's docs · read 2026-09-19
- Release date
- not published by the makermaker's docs · read 2026-09-19
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- KimiK3ForConditionalGeneration
- Layers
- 93
- Vocabulary
- 163,840 tokens
- Weights first published
- June 13, 2026
- Weights last updated
- September 2, 2026
moonshotai/Kimi-K3 · config.json @ f831ab6 · read 2026-10-08 · 6 of 10 fields published by the maker
| Date | Provider | Input | Output | Change |
|---|---|---|---|---|
| Oct 23 | Morph | $0.89 → $1.25 | $14.90 → $14.90 | +2% |
| Oct 17 | Parasail | $2.60 → $3.00 | $13.00 → $15.00 | +15% |
| Oct 14 | Together AI | $2.70 → $3.00 | $13.50 → $15.00 | +11% |
| Oct 13 | Chutes | $3.00 → $1.85 | $15.00 → $9.50 | −37% |
| Oct 7 | AkashML | $1.30 → $1.20 | $14.00 → $14.00 | −1% |
| Oct 7 | Parasail | $3.00 → $2.60 | $15.00 → $13.00 | −13% |
| Oct 6 | AkashML | $3.00 → $1.30 | $15.00 → $14.00 | −15% |
| Oct 2 | Together AI | $3.00 → $2.70 | $15.00 → $13.50 | −10% |
Every published change of input or output price, newest first · change on input + output · full series: pricing-history.csv.
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| Chutes | $6.60 | $19.55 | $16.90 |
| AkashML | $8.20 | $16.60 | $18.80 |
| Phala | $8.93 | $26.77 | $22.95 |
| Morph | $8.70 | $17.47 | $19.91 |
| Relace | $9.40 | $26.20 | $23.60 |
| DeepInfra | $9.97 | $29.93 | $25.65 |
| DigitalOcean | $10.50 | $31.50 | $27.00 |
| Parasail | $10.50 | $31.50 | $27.00 |
| Modal | $10.50 | $31.50 | $27.00 |
| Together AI | $10.50 | $31.50 | $27.00 |
| Fireworks AI | $10.50 | $31.50 | $27.00 |
| BaseTen | $10.50 | $31.50 | $27.00 |
| Moonshot AI | $10.50 | $31.50 | $27.00 |
| Novita AI | $10.50 | $31.50 | $27.00 |
| InferenceNet | $10.50 | $31.50 | $27.00 |
| Amazon Bedrock | $10.50 | $31.50 | $27.00 |
| Alibaba | $12.08 | $36.23 | $31.05 |
What each host bills for 1,000 requests of a representative job
Your own workload
Other Moonshot models · all its models, hosts and history ›
- Kimi K2.618 providers · from $2.40
- Kimi K2.7 Code15 providers · from $3.71
- Kimi K2.58 providers · from $2.70
- Kimi K2 Thinking2 providers · from $3.10
Head-to-head: Kimi K3 (Moonshot AI) vs DeepSeek V4 Pro 0813 (DeepSeek)