← All models

Kimi K3.1 Pro

Language Available Full comparison ↗

moonshot/kimi-k3.1-pro · by Moonshot AI · mixture-of-experts · Built on Kimi K3

Pricing — 1 offering(s)

Moonshot AI

About this offering: Kimi K3.1 Pro (API model id "kimi-k3.1-pro"), Moonshot AI's flagship for long-horizon coding and knowledge work, released October 2026. Moonshot describes it as a post-trained version of Kimi K3 and now calls K3 its previous generation. 1,048,576-token context window; text and image input. Flat pricing at every context length ($2.00 input / $10.00 output / $0.20 cached input per 1M tokens), a third cheaper than K3. Thinking is always on, with reasoning effort low, high or max (default max). Moonshot says the full model weights will be released by 4 November 2026.

Input tokens

  • $2.00 / 1M tokens (input) Current 2026-10-23 → present

Cached input tokens

  • $0.20 / 1M tokens (cached input) Current 2026-10-23 → present

Cache write tokens (5-minute TTL)

  • $2.00 / 1M tokens (cache write) Current 2026-10-23 → present
    5-minute TTL (Moonshot's default). Writes with a 1-hour TTL cost $4.00/1M. A cache hit refreshes the entry's…

    5-minute TTL (Moonshot's default). Writes with a 1-hour TTL cost $4.00/1M. A cache hit refreshes the entry's lifetime with no further write charge.

Output tokens

  • $10.00 / 1M tokens (output) Current 2026-10-23 → present
    Thinking is always on (reasoning_effort low / high / max, default max), and reasoning tokens are output…

    Thinking is always on (reasoning_effort low / high / max, default max), and reasoning tokens are output tokens, so a lower effort cuts output cost.

Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.

See how Kimi K3.1 Pro fits into a cost-aware routing setup

See how →

Capability profile

How Kimi K3.1 Pro rates across core capability dimensions, with the task-level evidence behind each rating.

coding Strong

Moonshot's flagship for long-horizon coding (docs index), with game-engine, 3D and front-end examples in the launch post. Vendor positioning; no independent coding benchmark recorded yet.

reasoning Strong

Always thinks before answering, with reasoning effort low, high or max (default max). Vendor positioning; no independent reasoning benchmark recorded yet.

tool use Strong

Supports custom tools with tool_choice, dynamic tool loading and Moonshot's official tools (K3.1 Pro quickstart). Moonshot says its web search tool is being updated and isn't recommended for production for now.

context window Strong

1,048,576-token context window per Moonshot's pricing page, with flat pricing at every context length. Output defaults to 131,072 tokens and can be raised to 1,048,576.

instruction following Unknown
multilingual Unknown
speed Unknown

No throughput figure published. Moonshot sells a separate Highspeed tier at twice the price that it says is 2-3x faster.

cost efficiency Moderate

$2.00/$10.00 per 1M tokens with cached input at $0.20: a third cheaper than Kimi K3 ($3/$15) and Moonshot's recommended upgrade from it.

Ratings are estimated — limited independent data is available for this model.

Operator guidance

Moonshot's current flagship, replacing Kimi K3 as the default: a post-trained K3 at a third lower price ($2/$10 per 1M tokens, flat up to 1M tokens). Thinking is always on and defaults to max effort, and reasoning tokens bill as output, so set reasoning_effort to low for simple tasks. Sampling parameters are fixed. It's also Moonshot's migration target for the K2 series (K2.6, K2.7 Code), which is being retired by the end of 2026. Capability claims are Moonshot's own so far, so check results on your task.

Use cases

Limitations

Citations