Kimi K3.1 Pro Highspeed API

Price check: official, OpenRouter and Wokey

USD per million tokens. Official and OpenRouter rates were checked by hand on 2026-10-24, excluding tax; the Wokey rate is read from the live price list.

Checked 2026-10-24 · Official pricing page · OpenRouter endpoint list

Use it in Claude Code

This model supports /v1/messages on Wokey. Set these variables, then start claude.

export ANTHROPIC_BASE_URL="https://api.wokey.ai"
export ANTHROPIC_AUTH_TOKEN="$WOKEY_API_KEY"
export ANTHROPIC_MODEL="kimi-k3.1-pro-highspeed"

Kimi K3.1 Pro Highspeed at a glance

Highspeed

The same model, which Moonshot says runs 2–3× faster. Wokey measured on 10-24 (Kimi Code endpoint, effort=high): Pro 64–67 tokens/s, Highspeed 118–129 tokens/s, about 1.9× faster; K3 ran about 45 tokens/s under the same conditions

Official price

Highspeed costs $4 input / $20 output per million tokens, twice Pro ($2 / $10); cache hits stay at $0.20, and cache writes are $4 (5-minute) or $8 (1-hour)

Release date

2026-10-23: K3.1 Pro and K3.1 Pro Highspeed shipped on the Kimi web app and mobile apps, Kimi Work, Kimi Code and the Kimi API on day one; Moonshot promises the full weights on Hugging Face by November 4

Thinking and parameters

Thinking-only: disabling thinking returns 400. reasoning_effort accepts low / high / max with max as the default; temperature, top_p and the other sampling parameters are fixed, the same constraints as K3

Independent evaluation

Artificial Analysis Intelligence Index v4.3.2: 55, ranked 4 of 215, behind only Claude Opus 5.5 (58), Sonnet 5.5 (56) and Muse Wonder Full (56), and 11 points above K3; $2.11 per task, about a third of Opus 5.5

API and channels

Model IDs are kimi-k3.1-pro and kimi-k3.1-pro-highspeed on the Kimi platform, and k3.1-pro and k3.1-pro-highspeed in Kimi Code. Wokey serves /v1/chat/completions and /v1/messages (works with Claude Code)

Checked 2026-10-24 · Sources: Moonshot · K3.1 Pro quickstart · Pricing · Kimi Code models · Artificial Analysis

Official benchmarks: Kimi K3.1 Pro vs Opus 5.5 vs GPT-6 Astra vs Fable 5.1 vs GLM-5.3 vs DeepSeek

From Moonshot’s K3.1 Pro tech blog; these are vendor-reported results, and “—” means no score was published. Highspeed is the same model, so the scores are identical. Internal and cybersecurity evaluations are left out.

Every K3.1 Pro score uses reasoning_effort=max, temperature=1.0 and top_p=0.95, with coding benchmarks in the Claude Code harness; the three proprietary models ran at max with fallback. The ProgramBench row is a binary score (a task counts when it reaches ≥95%), so it is not comparable with the raw scores on the K3 page. GLM-5.3’s HLE-Full score is on the text-only subset.

Which to use: Highspeed or Pro, and other models

Pro or Highspeed

They are the same model with the same benchmark scores. Highspeed costs twice as much and uses about twice the quota in Kimi Code; in Wokey’s measurements it is about 1.9× faster (118–129 vs 64–67 tokens/s) and reaches the first token about a second sooner. Choose Highspeed for interactive coding where someone waits on the answer; choose Pro for background agents, batch work and token-budgeted jobs.

vs Claude Opus 5.5

On Artificial Analysis’ Intelligence Index v4.3.2 Opus 5.5 (max) scores 58 and K3.1 Pro 55, only 3 points apart, while a task costs $5.98 versus $2.11, about a third for K3.1 Pro. Moonshot’s table has them nearly level on Terminal-Bench 4.0 (65.7% vs 66.4%), and tbench.ai’s cost data puts a task at about $5.25 for K3.1 Pro and $14.30 for Opus 5.5. K3.1 Pro leads on SWE-Marathon v1.1 (51.7% vs 46.3%), HLE with tools (70.0% vs 67.7%) and OfficeQA Pro v2; Opus 5.5 is clearly stronger on ProgramBench (65.0% vs 31.5%), Terminal-Bench Science (63.3% vs 38.1%) and HLE without tools (59.8% vs 51.1%), and faster (about 96 vs 77 tokens/s). Pick K3.1 Pro for long-horizon coding agents, document and finance analysis, and budget-sensitive volume; pick Opus 5.5 for complex programs that must work first time and for scientific work.

Upgrading from K3

Yes, for pay-per-token API use. K3.1 Pro scores 11 points higher than K3 on the index (55 vs 44), lists a third cheaper, and outputs about 64–67 tokens/s against K3’s 45 in Wokey’s measurements. The request constraints are the same as K3’s, so switching is a matter of changing model from kimi-k3 to kimi-k3.1-pro. Kimi recommends starting a new session: switching mid-way through a K3 session invalidates the cache and can lower output quality. Kimi Code subscribers should note that, in Wokey’s measurement on the Max plan, K3.1 Pro uses about 50% more quota on input and 82% more on output than K3 for the same tokens, so keep simple tasks on K3 when quota is tight.

Choosing effort

Thinking cannot be turned off, and reasoning_effort accepts only low, high and max, with max as the default. On the Terminal-Bench 4.0 cost curve Moonshot published, high scores 52.8% at about $4.16 per task and max 65.7% at about $5.25: max costs 26% more for 12.9 more points. Keep max for coding and agents and use low for simple Q&A and format conversion. Reasoning tokens bill at the output rate, and Artificial Analysis recorded 150M output tokens across its index, 70% above the 87M median, so leave room in output budgets.

Kimi docs

Third-party and community reviews

• Artificial Analysis: Intelligence Index v4.3.2: 55, ranked 4 of 215. It used 150M output tokens across the index (median 87M), costs $2.11 per task, outputs about 77 tokens/s and takes about 3.1 seconds to first token. It is listed as proprietary until the weights ship.
• Hacker News (542-point thread): The most discussed point was Moonshot removing cybersecurity material from the training data: some see it as sturdier than refusals, others worry about general coding skill. Next came Kimi’s subscription quota: one user on the $99 tier used up the 5-hour quota in 47 minutes of one session. Others noted the China price is about 30% below K3 and that it now ranks above GPT-6 Astra and Fable 5.1 on Artificial Analysis.
• Wokey measurements: On 10-24 Wokey ran the same prompt (a technical write-up of about 900 words) three times per model at the Kimi Code endpoint (effort=high): K3.1 Pro ran 64–67 tokens/s, Highspeed 118–129 tokens/s and K3 about 45 tokens/s. Both chat/completions and the messages format Claude Code uses return thinking normally; disabling thinking, top_p≠0.95 and public image URLs all return 400.

Using K3.1 Pro Highspeed in Claude Code, and request rules to check

These rules come from Kimi’s documentation and Wokey’s tests of both models on 10-24. Check them before switching from K3 or wiring it up for the first time.

• Model IDs: the Kimi platform and Wokey use kimi-k3.1-pro / kimi-k3.1-pro-highspeed; keys from a Kimi Code membership need k3.1-pro / k3.1-pro-highspeed. In Wokey’s test, sending kimi-k3.1-pro to Kimi Code returns 400 “Your model id does not exist”.
• Thinking cannot be turned off: thinking: {"type": "disabled"} returns 400 “K3.1 Pro does not support disabled thinking mode”. Use the top-level reasoning_effort (low / high / max) to control how much it thinks.
• temperature, top_p, n, presence_penalty and frequency_penalty are fixed at 1.0, 0.95, 1, 0 and 0, and Kimi asks you to omit them; top_p=1 returns 400 “invalid top_p: only 0.95 is allowed for this model”. The Wokey gateway enforces the same rules, so the answer does not depend on which upstream serves the request.
• In multi-turn chats and tool calls, pass back the complete assistant message the API returned, including reasoning_content and tool_calls.
• Images go in as base64 data URLs such as data:image/png;base64,.... In Wokey’s 10-24 test a public image URL returned 400 “unsupported image url”, as with K3; video also goes in as base64.
• In Claude Code: set ANTHROPIC_BASE_URL to https://api.wokey.ai, ANTHROPIC_AUTH_TOKEN to your Wokey API key and ANTHROPIC_MODEL to kimi-k3.1-pro-highspeed. Wokey tested both models on 10-24 in the /v1/messages format Claude Code uses, and thinking blocks and text came back normally; it costs $1.20 input / $6.00 output per million tokens. OpenCode and other OpenAI-compatible clients use /v1/chat/completions with the base URL https://api.wokey.ai/v1.

Request body with an image (/v1/chat/completions)

{
  "model": "kimi-k3.1-pro-highspeed",
  "reasoning_effort": "high",
  "messages": [{
    "role": "user",
    "content": [
      { "type": "image_url", "image_url": { "url": "data:image/png;base64,iVBORw0KGgo..." } },
      { "type": "text", "text": "Rebuild this CAD sketch as a parametric Blender script" }
    ]
  }]
}

Catalog pricing snapshot

Prices are per 1M tokens; request-time pricing controls settlement.

Official price source: https://platform.kimi.ai/docs/pricing/chat

Model capabilities

Context

1,048,576 tokens

Max output

1,048,576 tokens

Streaming

Supported

Tools

Supported

Vision

Supported

Supported API forms

/v1/chat/completions · /v1/messages

Send a request

/v1/chat/completions

curl https://api.wokey.ai/v1/chat/completions \
  -H "Authorization: Bearer $WOKEY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3.1-pro-highspeed","messages":[{"role":"user","content":"Hello"}]}'

/v1/messages

curl https://api.wokey.ai/v1/messages \
  -H "x-api-key: $WOKEY_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "Content-Type: application/json" \
  -d '{"model":"kimi-k3.1-pro-highspeed","max_tokens":1024,"messages":[{"role":"user","content":"Hello"}]}'

Use Kimi K3.1 Pro Highspeed in your tools

The gateway is https://api.wokey.ai (OpenAI-compatible clients usually take https://api.wokey.ai/v1), and the model ID is kimi-k3.1-pro-highspeed. Open the guide for the client you use.

• Use Kimi K3.1 Pro Highspeed in Claude Code: Anthropic Messages · base URL and API key setup
• Use Kimi K3.1 Pro Highspeed in Claude Desktop: Anthropic Messages · base URL and API key setup
• Use Kimi K3.1 Pro Highspeed in opencode: OpenAI Chat Completions · base URL and API key setup
• Use Kimi K3.1 Pro Highspeed in OpenClaw: OpenAI Chat Completions · base URL and API key setup
• Use Kimi K3.1 Pro Highspeed in Hermes Agent: OpenAI Chat Completions · base URL and API key setup
• Use Kimi K3.1 Pro Highspeed in DeepSeek Harness: OpenAI Chat Completions · base URL and API key setup
• Kimi Code limits, measured: how much Kimi K3.1 Pro Highspeed a plan covers: Measured tokens per 5-hour window and per month, and what they are worth at API prices.
• OpenRouter alternative: Wokey vs OpenRouter: Compare token prices, payment fees, API forms, and upstream sources.
• Verifiable AI API: which models and API forms carry response proofs: Response proofs currently cover only raw passthrough Claude Messages and GPT Responses on official routes.

Frequently asked questions

Related models and documentation

• Kimi K3.1 Pro
• Kimi K3
• Kimi K2.8 Preview
• API documentation
• All models and pricing
• Provider Node