Kimi K3.1 Pro launches with API rates one-third below K3

Moonshot AI has launched Kimi K3.1 Pro with standard API input and output rates one-third below those of K3. Kimi Code’s October 23 release notes confirm the rollout. The official API platform lists $2 per million input tokens and $10 per million output tokens, compared with K3’s $3 and $15.

The accompanying K3.1 Pro Highspeed version has API rates of $4 for input and $20 for output per million tokens, twice the standard rates. Actual request costs depend on input and generated volume, rather than the unit rate alone.

The company describes K3.1 Pro as a 3T-class, post-trained version of K3, with more reinforcement-learning compute and changes to long-context inference and execution environments. It reports 65.7% on Terminal-Bench 4.0, using the Claude Code harness. Unless otherwise specified, its reported results use maximum reasoning effort.

Independent evaluator Artificial Analysis lists an Intelligence Index score of 55, combining ten evaluations across areas including coding, science and document work. Its evaluation generated 150 million output tokens and recorded a weighted average cost of $2.11 per task. That figure describes the evaluation workload, rather than a fixed charge for an ordinary user request.

The model accepts text, images and video, calls external tools and supports answers constrained to a JSON schema. Thinking remains enabled at all times, with low, high and max effort levels. Requests that disable thinking are rejected; the service does not offer a mode that removes reasoning entirely for simpler responses.

The roughly one-million-token context window remains, although Kimi Code access depends on the subscription. Plus and legacy Moderato members get the standard model with up to 262,144 tokens. Pro and legacy Allegretto tiers or higher unlock 1,048,576 tokens and Highspeed. Using Highspeed in Code consumes approximately twice the standard model’s quota.

Code’s desktop, command-line, VS Code and third-party clients share subscription quota. New plans remove the weekly limit but retain a rolling five-hour rate window and a monthly total. Legacy plans keep their weekly limits as well. The membership documentation says Code quota freezes once the shared monthly total is reached.

The general Kimi API is billed separately from that subscription on a pay-as-you-go basis. Both input and output are charged; K3-series billing also distinguishes cache writes from cache hits. Extracted document contents count as input when sent to the model. Published prices exclude applicable taxes.

Moonshot says the model is available through Kimi’s web and mobile apps, Work, Code and API, with full weights due by November 4. Enterprise access is off by default at launch and can be enabled by organization administrators.