Kimi K3 Pricing 2026: Plans, API Costs, and What Each Tier Gets You

Kimi K3 pricing covers Kimi's four paid membership plans plus the developer API. K3 is Moonshot AI's 2.8-trillion-parameter flagship model for long-horizon reasoning, coding, and knowledge work. Kimi offers a free Adagio tier alongside paid plans starting at ¥49/month: Andante, Moderato, Allegretto, and Allegro. All membership features share a single credit pool. On the current ¥-priced plans, K3's 1-million-token context is available on Allegro, while Kimi Code is available across all paid tiers. The developer API is priced separately on a token basis.

Adagio: The Free Tier

Adagio gives you Kimi K3 for free with one active agent task at a time. Swarm (parallel subagents), Dream Memory, Goal, and Kimi Code are all locked on free. Scheduled and widget tasks cap at two. The core chat experience - research, writing, document work — is unrestricted. The limit shows the moment you need agentic workflows that run more than one task simultaneously.

Adagio does not include K3's extra-long chat capacity. The 1M-token context window is a paid-tier feature starting at Allegro. For a full ranking of free AI tiers across providers, see our free AI tools guide.

Moderato and Allegretto: Entry-Level Paid Plans

Moderato costs $19/month on monthly billing or $15/month on annual billing ($180/year billed upfront). It unlocks Swarm with two subagents, Dream Memory, Self-evolving Skill, website deployment, and the base tier of Kimi Code credits (1×). Agent priority queue runs at 4× speed versus the free tier. Scheduled tasks increase from 2 to 10.

Allegretto costs $39/month ($31/month annually, $372/year). It adds the Goal feature for long-horizon autonomous tasks, bumps Swarm to four subagents, increases Kimi Code credits to 5×, and unlocks Kimi Claw — the 24/7 autonomous web agent — plus group chat with Claw agents. Both Moderato and Allegretto cap at two concurrent agent tasks and do not include the 1M-token context window.

Allegro and Vivace: Full K3 Power

Allegro costs $99/month ($79/month annually, $948/year). It is the first tier to unlock K3 extra-long chat capacity: up to 1M tokens of context in a single conversation. Concurrent agent tasks double to four. Swarm scales to eight subagents. Kimi Code credits reach 15×. For large codebase work, deep-research threads, and document-heavy agent tasks, Allegro is the minimum tier that removes context limits.

Vivace costs $199/month ($159/month annually, $1,908/year). It matches Allegro on tasks, concurrency, and the 1M-token context window. The sole differentiator is Kimi Code credits, which jump to 30×. For teams running intensive coding pipelines via Kimi Code, that multiplier justifies the cost. Individual users rarely need to go past Allegro.

API Pricing

Kimi's API now offers multiple models, including Kimi K3, K2.7 Code, and K2.6. K3 remains the flagship model, with a 1M-token context window and pricing of $3 per million input tokens, $0.30 per million cached input tokens, and $15 per million output tokens. Kimi has also introduced K2.8 Preview for Kimi Code, offering up to 1M-token context and adjustable thinking levels.

Context caching cuts input costs by 90% — cached context drops from $3.00 to $0.30 per million tokens. For applications with large shared system prompts or repeated document context, caching is the primary cost lever. Output is always $15/1M regardless of caching. The full 1M-token context window is available to all API users — no subscription tier required.

For a side-by-side comparison of K3 API rates against GPT-5, Claude Fable, Gemini Flash, and Grok, see our AI API pricing guide.

The Catches

K3 always reasons. There is no cheaper non-thinking mode. If you are using the API for simple tasks - short classification, fast extraction, low-complexity generation - the $15/1M output rate is steep for work that does not benefit from chain-of-thought reasoning. Other providers offer faster, cheaper non-reasoning models for those use cases.

Monthly billing runs 20-27% more than annual. Moderato at $19 monthly versus $15 annual costs $48 more per year. At Vivace, that gap grows to $480 per year. There is no mid-term option: it is month-to-month or a full year, with no month-by-month billing after the annual purchase.

The 1M-token context wall sits at Allegro ($79/month annual). The two tiers below it - Moderato and Allegretto - do not include extra-long context regardless of price. If long-context chat is the reason you are paying, you are committing to at least $79/month on annual billing.

Kimi Code credits use a multiplier system (1×, 5×, 15×, 30×) with no public definition of the base unit. Before committing to a tier based on the credit multiplier, test Kimi Code on Moderato and measure your actual consumption rate against the base.

Which Plan Should You Pick

Start on Adagio (free) to test K3 for chat, research, and documents. One task at a time is enough for most casual use. Adagio is a proper free tier, not a limited trial.

Pick Moderato ($15/month annual) if you need Swarm, Dream Memory, and basic Kimi Code access. It is the minimum tier for agentic workflows.

Pick Allegretto ($31/month annual) if you use Goal tasks or need Kimi Claw for web automation. The four-subagent Swarm cap and 5× Kimi Code credits suit everyday power users who do not need the 1M context window.

Pick Allegro ($79/month annual) if the 1M-token context window is the feature you need. Large codebases, long research threads, and extended document sessions all benefit. This is the practical ceiling for most professionals.

Pick Vivace ($159/month annual) only if your Kimi Code usage is heavy enough that 30× credits pay off versus Allegro. For everything else, Allegro covers the same feature set at half the price.

Use the API for production tools or custom agents that need programmatic K3 access. The 1M context window is available at the API level without a subscription. Context caching keeps costs manageable on repeated-context workflows. For the full picture of what every major AI provider charges, see our AI pricing comparison.