Model
Kimi K3
Multimodal Kimi model with 1M context and toggleable max-effort thinking for long-horizon agent work
Compare providers (sorted by listed input price)
56 of 64 providers publish a price for Kimi K3. The cheapest full-context option is CrofAI at $2/$8 per M — 42% below Moonshot AI's own price. Listed prices span 2.6× for identical weights.
Best price history
Daily lowest published price across live providers. A lower line is better for the buyer.
- Best input /M
- $2unchanged
- Best output /M
- $8unchanged
- Provider coverage
- 78
- 66 at the start · 24 snapshots
Published best prices are unchanged.
24 daily snapshots collected from Aug 21, 2026 through Sep 14, 2026. A price chart will appear after the first real move.
Price badge
https://www.vaanalytics.in/badge/moonshotai/kimi-k3.svgLive SVG, regenerated on every hourly sync — paste it into a README to always show the current cheapest listed price. Updates when prices change; no build step needed on your side.
Recent changes
Changelog →- new modelvia llmgateway-providers · Sep 13, 2026
- new modelvia volcengine-coding-plan · Sep 13, 2026
- new modelvia llmgateway-providers · Sep 13, 2026
- new modelvia llmgateway-providers · Sep 13, 2026
- repricedvia openrouter · Sep 13, 2026
- input: $2.34 → $2.65
- output: $11.70 → $13.28
- cache_read: $0.261 → $0.303
- new modelvia pioneer · Sep 13, 2026
Benchmarks
- DeepSWE(resolve rate)report ↗Unclassified source67.5
- Terminal-Bench ↗(accuracy)report ↗Unclassified source88.3
- FrontierSWE ↗(dominance score)report ↗Unclassified source81.2
- Program Bench(score)report ↗Unclassified source77.8
- SWE Marathon ↗(resolve rate)report ↗Unclassified source42
- GDPval-AA(Elo)report ↗Unclassified source1668
- AA-Briefcase(Elo)report ↗Unclassified source1548
- AutomationBench(success rate)report ↗Unclassified source30.8
- JobBench(score)report ↗Unclassified source52.9
- SpreadsheetBench(score)report ↗Unclassified source34.8
- BrowseComp ↗(accuracy)report ↗Unclassified source91.2
- CharXiv Reasoning(accuracy)report ↗Unclassified source91.3
- ZeroBench(pass@5)report ↗Unclassified source41
Scores a lab published about its own model are claims, not measurements. Rankings elsewhere on this site use independently measured scores only.
Weights
No machine-readable licence on the model card. Check it before you rely on this. Weights download without an access request.
licence and gating from the Hugging Face model card ↗ — the card is authoritative, this is a summary
Prices are per million tokens (USD). “—” means the provider does not publicly list a price for this model. Data from models.dev; verify with the provider before purchasing.