KIMI FLAGSHIP1M CONTEXTTEXT + VISION + VIDEOALWAYS-ON THINKINGWEIGHTS PROMISED BY NOV 4
Kimi K3.1 Pro API Pricing
Kimi K3.1 Pro is Moonshot's flagship from October 24, 2026: a post-trained Kimi K3 with a 1M-token context, priced at $2.00/M input, $0.20/M cache-hit input and $10.00/M output. That is a third below K3 on every rate, and K3 stays on the price list as the previous generation. Thinking is always on, so output includes reasoning. Moonshot says the full weights will be released by November 4, 2026.
Input - per 1M tokens
$2.00/M
Source Kimi API -33% vs K3
Output - per 1M tokens
$10.00/M
Context 1M -33% vs K3
Cached input - per 1M tokens
$0.20/M
Cache automatic -90%
Effective - agentic blend
$1.28/M
92/8 split - 82% cache
§ 01 / TERMINAL
Kimi K3.1 Pro API cost calculator.
Live calculator pre-loaded with Kimi K3.1 Pro rates. Reasoning effort runs low, high or max (the default), and reasoning tokens bill as output, so put them in the reasoning field.
$ /mo
Workload split
Prompt cache hit rate
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
§ 02 / SCENARIOS
Kimi K3.1 Pro cost on real workloads.
CODING AGENT
Codebase-scale migration
$0.163/task
LONG DOC
Reading 100-page contracts
$0.174/doc
RAG SUPPORT
Ticket triage
$0.014/ticket
ASSISTANT
Research planning turn
$0.029/turn
Estimate · moonshot-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
Characters 653
Words 107
Tokens (estimated) 170 tokens
Cost as input · uncached $0.00034 USD
Cost as output · uncached $0.0017 USD
Cost as cached input $0.00003 USD
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Kimi K3.1 Pro Current | $2.00 cache $0.20 | $10.00 | $1.28 agentic 92/8 | 1M | Frontier Kimi agents - 1M context |
| Kimi K3 | $3.00 cache $0.30 | $15.00 | $1.92 +50% | 1M | The previous Kimi flagship |
| GPT-6.1 Sol | $2.00 cache $0.10 | $10.00 | $1.29 same | 1.05M | OpenAI's model at the same list price |
| Claude Opus 5.5 | $4.00 cache $0.20 | $20.00 | $2.58 +2.0× | 1M | Frontier coding agents |
| GLM-5.3 | $1.40 cache $0.26 | $4.40 | $0.78 −39% | 1M | Cheaper Chinese frontier, 1M context |
| DeepSeek V4 Pro | $1.32 cache $0.044 | $3.96 | $0.569 −2.3× | 1M | Budget Chinese frontier |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 −2.7× · promo | 1M | Long-context multimodal work |
Kimi K3.1 Pro has published scores on 17 benchmarks we track, and leads 9 of them. All of them are the lab's own reports; each row links to the page that published it.
| Benchmark | Score | Rank | Leader | Source |
|---|---|---|---|---|
| Terminal-Bench 4.0 Agents and computer use | 65.7% max effort | #1 of 33 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| Humanity's Last Exam Expert knowledge · With tools | 70% max effort, with tools | #1 of 19 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| CharXiv Reasoning Vision and documents · No tools | 89% max effort, no tools | #1 of 9 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| DeepSWE Coding | 75.5% max effort | #3 of 27 | Gemini 4 Argon · 77.9% | Vendor-reported Moonshot AI |
| Harvey Legal Agent Benchmark Professional work | 23.7% max effort | #1 of 7 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| ProgramBench Coding | 31.5% max effort | #1 of 7 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| Agents' Last Exam Agents and computer use · CLI split | 44.2% max effort, CLI split | #1 of 6 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| APEX-Agents Professional work | 69.4% max effort | #1 of 6 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| BrowseComp Agents and computer use | 92.1% max effort | #1 of 6 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| Chartography Vision and documents · With tools | 90.2% max effort, with Python | #1 of 5 | Kimi K3.1 Pro leads | Vendor-reported Moonshot AI |
| GDP.pdf Vision and documents · Default setting | 34.6% max effort, no tools | #2 of 7 | Gemini 3.8 Flash · 35% | Vendor-reported Moonshot AI |
| GPQA Diamond Expert knowledge | 94.2% max effort | #4 of 11 | GPT-6 Astra · 96% | Vendor-reported Moonshot AI |
| ExploitGym Cybersecurity | 20.7% max effort | #3 of 8 | GPT-6 Astra · 42.4% | Vendor-reported Moonshot AI |
| ExploitBench Cybersecurity | 74.7% max effort | #3 of 7 | GPT-6 Astra · 100% | Vendor-reported Moonshot AI |
| OSWorld 2 Agents and computer use · 2.1, partial score | 79.6% max effort | #4 of 6 | Claude Opus 5.5 · 81.8% | Vendor-reported Moonshot AI |
| Terminal-Bench-Science AI research | 38.1% max effort | #6 of 9 | GPT-6 Astra · 68.1% | Vendor-reported Moonshot AI |
| PostTrainBench AI research | 36.5% max effort, v1.2 | #5 of 6 | Gemini 4 Argon · 45.3% | Vendor-reported Moonshot AI |
Kimi K3.1 Pro API pricing FAQ.
Kimi K3.1 Pro pricing questions - the card, what changed from Kimi K3, caching and the Highspeed variant.
Q · 01 What is Kimi K3.1 Pro priced at? +
Moonshot's pricing page lists
$2.00/M cache-miss input, $0.20/M cache-hit input and $10.00/M output, with cache writes at $2.00/M for the 5-minute TTL and $4.00/M for 1 hour, for a 1,048,576-token context window. Under AI//COST's 92/8 agentic blend with 82% cache hits, the effective planning figure is $1.28/M. Prices exclude local taxes.Q · 02 What changed from Kimi K3? +
Kimi K3 lists at
$3.00/M input and $15.00/M output and blends to $1.92/M; K3.1 Pro blends to $1.28/M, with every rate a third lower. Moonshot describes K3.1 Pro as a post-trained K3 with substantially larger post-training compute, and reasoning effort now offers low and high as well as max, where K3 accepted only max. K3 stays on the price list with no deprecation date.Q · 03 What does Kimi K3.1 Pro Highspeed cost? +
kimi-k3.1-pro-highspeed, which Moonshot says runs at 2-3x the speed of K3.1 Pro, lists at $4.00/M input, $0.20/M cache-hit input, $4.00/M and $8.00/M cache writes, and $20.00/M output - twice the standard rates except the cache read. This page prices the standard model.Q · 04 How does prompt caching work? +
Automatically. Kimi writes eligible request prefixes with a 5-minute TTL by default, billed at
$2.00/M - the same as fresh input, so a miss costs nothing extra - or with a 1-hour TTL at $4.00/M when you ask for it. Cache hits bill at $0.20/M, a 90% discount, and the calculator's default blend assumes 82% of input is served from cache.Q · 05 How does it compare with GPT-6.1 Sol? +
GPT-6.1 Sol has the same
$2.00 / $10.00 sticker but reads cache at $0.10/M, so it blends to $1.29/M against $1.28/M here. GPT-6.1 Sol also moves to a dearer card above 272K input tokens; Kimi K3.1 Pro has one rate across its 1M context.Q · 06 Are the weights open? +
Not yet. Moonshot's launch post says the full model weights will be released by November 4, 2026, as they were for Kimi K3, which Moonshot calls the first open-source 3T-class model. Until then the model is available only on Kimi.com, Kimi Work, Kimi Code and the Kimi API, which offers Zero Data Retention to API customers.
Q · 07 What happens to the Kimi K2 models? +
Moonshot closes the K2 series - Kimi K2.6 and Kimi K2.7 Code - to newly registered users from October 31, 2026, keeps them for existing users, and plans to retire them by the end of December 2026. Its migration guide points to K3.1 Pro.
Q · 08 How accurate is the tokenizer estimate? +
The browser widget uses a
moonshot-tokenizer-estimate chars-per-token estimate for English text. Actual billing comes from Kimi API usage fields and can differ for Chinese, code, images, video, or mixed-language prompts.