Kimi K2 Thinking
MoonshotReleased Nov 6, 2025Open weights · Modified MITkimi-k2-thinking
- Intelligence (Epoch)
- 146 · 59th of 111
- Reasoning
- yes
- Knowledge cutoff
- April 2026
- Context
- 262K tokens
- Max output
- 100k tokens
- Input
- text
- API features
- tools, structured outputs, JSON mode, seed, logprobs
Read Oct 23, 06:00 UTC from the hosts' published offers and OpenRouter, spec sheet from Moonshot · cutoff: aiknowledgecutoff.com
The cheapest host, Google, does not state its weights precision. Cheapest at the maker's precision and context: Novita AI, $3.10 blended. Weights and context per host
Intelligence for the price
17 models score higher for the same price or less; it ranks 18 of the 68 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, 146 with an interval of 143 to 148.
Against its rivals
The best of its price range and the top score
- Kimi K2 Thinkingthis model$3.10146
- Claude Opus 5.5top score$24.00169
- Gemini 3.7 Flashsame price range$4.50158
- Gemini 3.8 Flashsame price range$4.50157
Same level for 19× less: Qwen3.7 Flash at Alibaba, $0.16 per M input + output · their Epoch intervals overlap (144 against 146).
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.
Run it yourself
open weights, Modified MIT · too large for the 6 workstations in our guideSelf-host guide ›- Parameters
- 1000B
- 32B active per token (MoE)
- Memory needed
- 600 GB
- in INT4 · 1200 GB in INT8
- Smallest cloud setup
- 8× H100
- 640 GB · $17.20/h, about $12,556 a month
- Worth it above
- 4B tokens
- a month, against the cheapest API at $3.10 per M
| Quantization | Memory | Smallest cloud GPU | Quality |
|---|---|---|---|
| INT4 (Q4) | 600 GB | 8× H100 · $17.20/h | smallest footprint, minor quality loss |
| INT8 (Q8) | 1200 GB | multi-node, over 640 GB | near-lossless |
| FP16 (full) | 2400 GB | multi-node, over 640 GB | reference quality |
At home: 600 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 ↑ | Uptime 24h | Checked |
|---|---|---|---|---|---|
| Novita AIbf16 · 256K ctx · 96K out | $0.60 | $2.50 | $0.15 | — | Oct 23 |
| Googleprecision not stated · 256K ctx | $0.60 | $2.50 | — | 84.3% | Oct 22 |
2 of 2 providers · prices per million tokens, read Oct 23, 06:00 UTC · uptime: OpenRouter, last 24 h · 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 |
|---|---|---|---|---|---|---|
| not stated by the host | 256K | not stated by the host | yes | yes | yes | |
| Novita AI | bf16 | 256K | 96K | 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 K2 Thinking spec sheet, field by field, with sources
Kimi K2 Thinking 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
- 262,144 tokens
- Parameters
- 1,026,408,240,256≈ 1.03T
- Active experts
- 8 of 384 active per token
- Weight precision
- compressed-tensors
- 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
- Max output (official)
- to verify
- Release date
- to verify≈ November 6, 2025 third-party catalogue — not the maker's figure
- Maker lifecycle status
- to verify
Also declared in the weights
- Architecture class
- DeepseekV3ForCausalLM
- Layers
- 61
- Vocabulary
- 163,840 tokens
- Weights first published
- November 4, 2025
- Weights last updated
- January 30, 2026
moonshotai/Kimi-K2-Thinking · config.json @ a51ccc0 · read 2026-10-08 · 6 of 10 fields published by the maker
| Provider | Chatbot 1,000 in / 500 out | RAG / long context 8,000 in / 500 out | Batch summarize 4,000 in / 1,000 out |
|---|---|---|---|
| $1.85 | $6.05 | $4.90 | |
| Novita AI | $1.85 | $6.05 | $4.90 |
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 K317 providers · from $11.35
- Kimi K2.7 Code15 providers · from $3.71
- Kimi K2.58 providers · from $2.70