Kimi K3
- In Input modalities: TextImageVideo Out Output modalities: Text
- Reasoning· low, high, max
- Reasoning can be off
- Tool use
- Structured output
- Open weights
Prices
10 offerings · prices as listed| Conditions/ 1M tokens | Input | Cache read | Cache write | Output | |
|---|---|---|---|---|---|
| 5-minute cache | 1-hour cache | ||||
| Standard | $3.00 | $0.30 | $3.00 | $6.00 | $15.00 |
Prices from Moonshot AI platform pricing (opens in a new tab) (checked Nov 8, 2026)
| Conditions/ 1M tokens | Input | Cache read | Cache write | Output | |
|---|---|---|---|---|---|
| 5-minute cache | 1-hour cache | ||||
| Standard | ¥20.00 | ¥2.00 | ¥20.00 | ¥40.00 | ¥100.00 |
Prices from Kimi API 开放平台：模型推理价格 (opens in a new tab) (checked Nov 8, 2026)
| Conditions/ 1M tokens | Input | Cache read | Output |
|---|---|---|---|
| Standard | $2.827 | $0.283 | $14.133 |
Prices from Alibaba Cloud Model Studio pricing (opens in a new tab) (checked Nov 9, 2026) · Kimi K3 model (opens in a new tab) (checked Nov 9, 2026)
| Conditions/ 1M tokens | Input | Cache read | Output |
|---|---|---|---|
| Standard | $3.00 | $0.30 | $15.00 |
Prices from Alibaba Cloud Model Studio pricing (opens in a new tab) (checked Nov 9, 2026) · Kimi K3 model (opens in a new tab) (checked Nov 9, 2026)
| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard | ¥20.00 | ¥100.00 |
Prices from 阿里云百炼：模型价格 (opens in a new tab) (checked Nov 9, 2026)
kimi/kimi-k3| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard | ¥20.00 | ¥100.00 |
Prices from 阿里云百炼：模型价格 (opens in a new tab) (checked Nov 9, 2026)
| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard no longer listed since Oct 23, 2026 | $2.827 | $14.133 |
Prices from Alibaba Cloud Model Studio pricing (opens in a new tab) (checked Nov 9, 2026)
| Conditions/ 1M tokens | Input | Output |
|---|---|---|
| Standard | $3.00 | $15.00 |
Prices from Alibaba Cloud Model Studio pricing (opens in a new tab) (checked Nov 9, 2026)
Standard: the price a request with no conditions pays (the headline price elsewhere on the site).
Details
- Licence
- Kimi K3 License
- Family
- kimi-k
- Model ID
Not stated on Moonshot AI’s pages: max input, max thinking budget, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: moonshotai/Kimi-K3- Parameters 2.78T
- Active not stated · 896 experts, 16 per token
- Native precision BF16
- Layers 93
- KV cache per token 105 KB
- Config positions 1024K (config.json max_position_embeddings)
- API context 1024K (maker’s API)
- Needs
- ≈ 1445.0 GB (weights 1375.4 + KV cache 0.8 + overhead 68.8)
- Usable
- 21.6 GB (90% of its memory)
- Try
- it’s too large for this hardware; see the smallest that runs it below
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 1453.7 GB | ≈ 1527.2 | ✕Doesn’t fit | — |
| Q8_0 | 2750.8 GB | ≈ 2889.2 | ✕Doesn’t fit | — |
| FP8 | 2589.0 GB | ≈ 2719.3 | ✕Doesn’t fit | — |
| Q6_K | 2126.2 GB | ≈ 2233.4 | ✕Doesn’t fit | — |
| Q5_K_M | 1841.4 GB | ≈ 1934.3 | ✕Doesn’t fit | — |
| Q5_0 | 1779.9 GB | ≈ 1869.8 | ✕Doesn’t fit | — |
| Q4_K_M | 1572.8 GB | ≈ 1652.3 | ✕Doesn’t fit | — |
| Q4_0 | 1456.3 GB | ≈ 1530.0 | ✕Doesn’t fit | — |
| IQ4_XS | 1375.4 GB | ≈ 1445.0 | ✕Doesn’t fit | — |
| Q3_K_Mquality drops | 1265.4 GB | ≈ 1329.5 | ✕Doesn’t fit | — |
| IQ3_XXSquality drops | 990.3 GB | ≈ 1040.6 | ✕Doesn’t fit | — |
| Q2_Kquality drops | 970.9 GB | ≈ 1020.2 | ✕Doesn’t fit | — |
| IQ2_XXSquality drops | 666.7 GB | ≈ 700.8 | ✕Doesn’t fit | — |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): no single listed machine; about 23 × 80 GB GPUs.
Run commands
The maker publishes no files at this precision. Choose its native precision or one of its own builds, or quantise it yourself with llama.cpp.
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-11. How it’s estimated
Cited scores
From third-party leaderboards, not measured by us- Epoch Capabilities Index 158.5
#21 of 118 listed models · CI 156.5–160.6
- LMArena Text 1,487
#15 of 163 listed models · max effort · CI 1,483–1,492 · 40,409 votes
- Epoch AI · Chess Puzzles 37%
#38 of 85 listed models · high effort · ±4.85
Also listed: max effort 36% · low effort 18%
- Epoch AI · FrontierMath Tier 4 v2 39%
#26 of 65 listed models · max effort · ±7.71
- Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Oct 28 – Nov 5, 2026
- LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Nov 8, 2026
- EQ-Bench by Samuel J. Paech (eqbench.com; EQ-bench/EQ-bench-site on GitHub), MIT License. · License · as of Oct 24 – 28, 2026
- LiveBench (White et al.), release 2026-06-25, livebench.ai, licensed CC BY-SA 4.0. This board, adapted from it, is shared under CC BY-SA 4.0. · License · as of Jun 25, 2026
- changes we made
Provider-reported scores
Published by Moonshot AI, under the conditions it states| Benchmark | Setting | Score |
|---|---|---|
| GPQA Diamond | reasoning effort max | 93.5% |
| Humanity's Last Exam | full set, no tools, reasoning effort max | 43.5% |
| full set, with general tools, reasoning effort max | 56% | |
| Terminal-Bench 2.1 | Kimi Code harness, reasoning effort max | 88.3% |
| BrowseComp | context compaction at 300K tokens, reasoning effort max | 91.2% |
| full 1M-token context, no context management, reasoning effort max | 90.4% | |
| MCP-Atlas | 500-task public subset, 100-turn limit, Gemini 3.1 Pro judge, reasoning effort max | 84.2% |
| OSWorld-Verified | reasoning effort max | 84.8% |
| MMMU-Pro | no tools, avg of 3 runs, reasoning effort max | 81.6% |
| with Python tool, avg of 3 runs, reasoning effort max | 83.4% |
Included in coding plans
- 5× the usage of the Lite plan
- GitHub Copilot Pro $10.00 / month1,500 credits per monthVariable additional usage on top of your base. Flex allotments may change over time.15 USD of usage per month
- GitHub Copilot Pro+ $39.00 / month7,000 credits per monthVariable additional usage on top of your base. Flex allotments may change over time.70 USD of usage per month
- GitHub Copilot Max $100.00 / month20,000 credits per monthVariable additional usage on top of your base. Flex allotments may change over time.
- GitHub Copilot Business $19.00 / month1,900 credits per month
- GitHub Copilot Enterprise $39.00 / month3,900 credits per month
- Kimi Code (Kimi membership) Plus ¥99.00 / monthFor new members, the weekly quota limit is removed, and only the rolling 5-hour rate window remains.Both groups share the Kimi membership monthly total quota
- Kimi Code (Kimi membership) Pro ¥199.00 / monthFor new members, the weekly quota limit is removed, and only the rolling 5-hour rate window remains.Both groups share the Kimi membership monthly total quota
- Kimi Code (Kimi membership) Max ¥699.00 / monthFor new members, the weekly quota limit is removed, and only the rolling 5-hour rate window remains.Both groups share the Kimi membership monthly total quota
- Kiro Pro $20.00 / month1,000 credits per month
- Kiro Pro+ $40.00 / month2,000 credits per month
- Kiro Pro Max $100.00 / month5,000 credits per month
- Kiro Power $200.00 / month10,000 credits per month
News
All news →- Oct 31, 2026Capability Moonshot AI adds Kimi K3.1 Pro and K3 to Batch API Batch API now supports kimi-k3.1-pro and kimi-k3 on both the CN and Global platforms. Batch inference costs 60% of standard rates; these models also incur Cache Write fees, and cache hits are best-effort.
Changes
All changes →- Now priced on Alibaba Cloud Model Studio, China Input ¥20.00 · Output ¥100.00
- Now priced on Alibaba Cloud Model Studio, China Input ¥20.00 · Output ¥100.00
- 1 new price on Alibaba Cloud Model Studio, International Cache read $0.30
- 2 new prices on Moonshot AI Platform Cache write (5-minute cache) $3.00 · Cache write (1-hour cache) $6.00
Sources
- moonshotai/Kimi-K3 model card (opens in a new tab) huggingface.co · checked Nov 6, 2026
- Moonshot AI platform pricing (opens in a new tab) platform.moonshot.ai · checked Nov 8, 2026
- Kimi API 开放平台：模型推理价格in Chinese (opens in a new tab) platform.moonshot.cn · checked Nov 8, 2026
- Kimi API models overview (opens in a new tab) platform.kimi.ai · checked Nov 8, 2026
- Kimi API structured output (opens in a new tab) platform.kimi.ai · checked Nov 8, 2026
- Kimi API chat completions reference (opens in a new tab) platform.kimi.ai · checked Nov 8, 2026
- Kimi API responses reference (opens in a new tab) platform.kimi.ai · checked Nov 8, 2026
- Kimi platform changelog (opens in a new tab) platform.kimi.ai · checked Nov 9, 2026
- models.dev: kimi-k3 (community data, MIT) (opens in a new tab) github.com · checked Sep 30, 2026
- Kimi K3 - Kimi API Platform (opens in a new tab) platform.kimi.ai · checked Nov 4, 2026
- Kimi K3 model (opens in a new tab) alibabacloud.com · checked Nov 9, 2026
- Explore Qwen3, Wan2.6 & Latest Models on Model Studio - Alibaba Cloud Model Studio - Alibaba Cloud_Alibaba Cloud Model Studio(Model Studio) - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center (opens in a new tab) alibabacloud.com · checked Nov 9, 2026
- 阿里云百炼：模型价格in Chinese (opens in a new tab) help.aliyun.com · checked Nov 9, 2026
- 模型上下架与更新-大模型服务平台百炼-阿里云_大模型服务平台百炼(Model Studio)-大模型服务平台百炼(Model Studio)-阿里云帮助中心in Chinese (opens in a new tab) help.aliyun.com · checked Nov 4, 2026
- moonshotai/Kimi-K3 model card (opens in a new tab) huggingface.co · checked Nov 3, 2026
- Kimi K3: 2.8T Open Model for Coding & Knowledge Work (opens in a new tab) kimi.ai · checked Nov 9, 2026
- moonshotai/Kimi-K3 config.json (opens in a new tab) huggingface.co · checked Nov 4, 2026
- moonshotai/Kimi-K3 files (opens in a new tab) huggingface.co · checked Nov 6, 2026
- Alibaba Cloud Model Studio pricing (opens in a new tab) alibabacloud.com · checked Nov 9, 2026
- Text generation models - Qwen, DeepSeek, GLM | Model Studio - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center (opens in a new tab) docs.modelstudio.console.alibabacloud.com · checked Nov 8, 2026
- Alibaba Cloud Model Studio pricing (opens in a new tab) alibabacloud.com · checked Nov 9, 2026
Moonshot AI’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Moonshot AI · China · moonshot.ai
Related models
Snapshots, the same family and other models from this provider
- Kimi K2.5
- kimi-k2-0905-preview
- kimi-k2-0711-preview
- Kimi K3.1 Pro
- Kimi K3.1 Pro Highspeed
- Kimi-K2-Thinking
- kimi-k2-turbo-preview
- Moonshot-Kimi-K2-Instruct
- Claude Opus 5.5 vs Kimi K3
- Gemini 3.1 Pro preview vs Kimi K3
- Gemini 4 Argon vs Kimi K3
- GLM-5.3 vs Kimi K3
- GPT-6 Astra vs Kimi K3
- Grok 4.7 vs Kimi K3
- Kimi K3 vs MiniMax-M3
- Gemini 3.7 Flash vs Kimi K3