Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work
Kimi K3 by moonshotai is offered by 88 providers on this page. Public prices are shown for 80 of them. Its official list price is $3.00 per 1M input tokens and $15.00 per 1M output tokens. The lowest paid channel is CrofAI at $2.00 / $8.00 per 1M, about 9.0× below the list price. That channel is a third-party gateway, so confirm its availability and rate limits before depending on it. It also has 3 free ($0) channels and 5 covered by a paid subscription; free tiers usually carry rate limits, and subscription-covered access bills $0 per token only after the subscription fee.
Artificial Analysis rates it 43.6 on the Intelligence Index, with 76.2 for coding and 50 for agentic tasks. Against its lowest blended price of $3.50 per 1M, that is roughly 12 index points per dollar, which is the value ratio the leaderboards rank on. Median output speed is 44 tokens per second, with 3.74s to the first token. Running one task of the Artificial Analysis suite costs about $2.0001, which reflects how many tokens its reasoning consumes rather than the unit price alone.
The context window is 1,048,576 tokens at the reference host, but hosts report different limits, from 262,144 to 1,113,088, so the usable window depends on the provider you pick. It supports reasoning, tool use, and structured output. Accepted input modalities are Text, Image, Video, and PDF. The weights are open, so it can also be self-hosted or served through a gateway of your choice. Providers report the capability flags inconsistently, so verify a specific feature against the host you plan to use.
Quality is independently evaluated by Artificial Analysis. Speed/latency are model-level medians.
| Provider | Tier | Input | Output | Cache read | Cache write | Context | Output limit | Status |
|---|---|---|---|---|---|---|---|---|
| SenseNova (China) | Gateway | Free | $0 | — | 1,048,576 | 65,536 | ||
| Umans AI Coding Plan umans-kimi-k3 | Gateway | Subscription | $0 | $0 | 1,048,576 | 131,072 | ||
| Volcengine Ark Coding Plan | Gateway | Subscription | $0 | — | 1,048,576 | 131,072 | ||
| Kimi For Coding (kimi.com) k3 | Gateway | Subscription | $0 | $0 | 1,048,576 | 131,072 | ||
| SCNet Token Plan Kimi-K3 | Gateway | Subscription | $0 | — | 1,048,576 | 131,072 | ||
| Nvidia | Cloud | Free | — | — | 1,048,576 | 131,072 | ||
| Kenari | Gateway | Free | — | — | 1,048,576 | 131,072 | ||
| Kimi For Coding (kimi.ai) k3 | Gateway | Subscription | $0 | $0 | 1,048,576 | 131,072 | ||
| CrofAI | Gateway | $2.00 | $8.00 | $0.25 | — | 1,000,000 ⚠ | 262,144 | |
| engy | Gateway | $1.95 | $9.75 | $0.20 | — | 1,113,088 ⚠ | 65,536 | |
| Moonshot AI (China)Official | First-party | $3.00 | $15.00 | $0.30 | — | 1,048,576 | 1,048,576 | |
Sorted by blended price (input×0.75 + output×0.25) asc. The official channel always shows regardless of rank. Whether a gateway's low price is actually usable can't be verified.
Interleaved thinking (reasoning between tool calls) is declared by 43 of 88 providers.
7 / 88 providers expose no reasoning control (reasoning_options: []).
| Mode | Provider | Input | Output | Cache read | Cache write |
|---|---|---|---|---|---|
| priority | Fireworks AI | $3.75 | $18.75 | $0.38 | — |
8 more channels offer $0 (SenseNova (China), Umans AI Coding Plan, Volcengine Ark Coding Plan etc.); free tiers usually have rate limits and no SLA, excluded from ranking.
| Name | Conditions | Score | Metric | Source |
|---|---|---|---|---|
| DeepSWE | harness: Kimi Code · variant: max effort · v1.1 | 67.5 | resolve rate | Source ↗ |
| Terminal-Bench | harness: Kimi Code · variant: max effort · v2.1 | 88.3 | accuracy | Source ↗ |
| FrontierSWE | harness: Kimi Code · variant: max effort | 81.2 | dominance score | Source ↗ |
| Program Bench | harness: Kimi Code · variant: max effort | 77.8 | score | Source ↗ |
| SWE Marathon | harness: Claude Code · variant: max effort · v1.1 | 42 | resolve rate | Source ↗ |
| GDPval-AA | variant: max effort · vv2 | 1668 | Elo | Source ↗ |
| AA-Briefcase | variant: max effort | 1548 | Elo | Source ↗ |
| AutomationBench | variant: max effort · dataset: 600-task public subset | 30.8 | success rate | Source ↗ |
| JobBench | variant: max effort | 52.9 | score | Source ↗ |
| SpreadsheetBench | harness: Claude Code · variant: max effort · v2 | 34.8 | score | Source ↗ |
| BrowseComp | variant: max effort, context compaction | 91.2 | accuracy | Source ↗ |
| CharXiv Reasoning | variant: max effort, with tools | 91.3 | accuracy | Source ↗ |
| ZeroBench | variant: max effort, with tools | 41 | pass@5 | Source ↗ |
The same benchmark scores very differently across harness / dataset, so the qualifying conditions must be shown together.
Individual evaluations run by Artificial Analysis, on the same reasoning tier as the intelligence score above. Each benchmark has its own task set and harness, so rows are not comparable with one another. The Intelligence Index above draws on a different, newer set of evaluations.