Kimi K3.1 Pro vs Kimi K3
Kimi K3.1 Pro scores higher on the Models at Work leaderboard (86 vs 69 out of 100). Kimi K3.1 Pro is about 1.5× cheaper per blended million tokens ($4 vs $6). On Artificial Analysis' independent Intelligence Index, Kimi K3.1 Pro leads 55 to 44.
| Kimi K3.1 Pro · Moonshot | Kimi K3 · Moonshot | |
|---|---|---|
| Leaderboard rank | #2 | #18 |
| Score (0–100) | 86 | 69 |
| Tier | Open weights | Open weights |
| Price per 1M tokens | $2 in · $10 out | $3 in · $15 out |
| Context window | 1M | Not verified |
| Open weights | Yes | Yes |
| Artificial Analysis Intelligence Index | 55 | 44 |
| Terminal-Bench 4.0 (vendor) | 65.7% | — |
| Arena text score | — | 1488 |
| Best for | Agentic coding with an open-weight exit, 1M-token long-context work, Moving off Kimi K3 | Self-hosting, Open-licence requirements, Arena-style chat quality |
Compare source-qualified API rates and a monthly workload example before estimating your bill.
Choose Kimi K3.1 Pro if…
- you need agentic coding with an open-weight exit
- you need 1m-token long-context work
- you need moving off kimi k3
Fourth on the Artificial Analysis index at 55, one point behind Sonnet 5.5, at the same $2 in / $10 out list price, with the weights still to come.
Caveats: Scores within three points of Sonnet 5.5 (tie), and ranks on evals and price alone because no practitioner sentiment exists yet. Moonshot says the full weights arrive by 4 Nov 2026; until then it is open-weight in intent, hosted-only in practice. Artificial Analysis rates it slow (51 tokens/s) and very verbose: 150M tokens to run its index, which is where the $2.11 per-task cost comes from. Moonshot's Terminal-Bench figure is vendor-run and is not counted. No Arena or Scale result yet, and no practitioner threads in Signals, so sentiment is left out.
Choose Kimi K3 if…
- you need self-hosting
- you need open-licence requirements
- you need arena-style chat quality
The open-licence model that topped Arena earlier this month, now the previous generation: Moonshot priced it at $3 in / $15 out and has already shipped its successor.
Caveats: Superseded by Kimi K3.1 Pro (23 Oct); Artificial Analysis no longer lists K3 on its current leaderboard, so its index (44) and Arena figures are from 8-10 Oct. Context window not verified from a primary source.
What practitioners say
Kimi K3.1 Pro: No practitioner threads recorded in Signals yet; left out of the score rather than guessed.
Questions people ask
Which is better, Kimi K3.1 Pro or Kimi K3?
Kimi K3.1 Pro scores higher on the Models at Work leaderboard (86 vs 69 out of 100). Kimi K3.1 Pro is about 1.5× cheaper per blended million tokens ($4 vs $6). On Artificial Analysis' independent Intelligence Index, Kimi K3.1 Pro leads 55 to 44.
When should I choose Kimi K3.1 Pro over Kimi K3?
Choose Kimi K3.1 Pro for agentic coding with an open-weight exit, 1m-token long-context work, moving off kimi k3. Fourth on the Artificial Analysis index at 55, one point behind Sonnet 5.5, at the same $2 in / $10 out list price, with the weights still to come.
When should I choose Kimi K3 over Kimi K3.1 Pro?
Choose Kimi K3 for self-hosting, open-licence requirements, arena-style chat quality. The open-licence model that topped Arena earlier this month, now the previous generation: Moonshot priced it at $3 in / $15 out and has already shipped its successor.
Which is cheaper, Kimi K3.1 Pro or Kimi K3?
Kimi K3.1 Pro costs $2 per million input tokens and $10 per million output tokens on Moonshot's list price (Kimi API model pricing, read Oct 26, 2026). Kimi K3 costs $3 per million input tokens and $15 per million output tokens on Moonshot's list price (Kimi API model pricing, read Oct 26, 2026).
Scores come from the Models at Work leaderboard: Score is 0 to 100 and computed, not typed: 55% independent evals (Artificial Analysis, Arena, Scale SEAL, Epoch, each scaled against its natural floor and the best score in this table, then averaged), 15% blended price on a fixed log scale ($0.05 per million is 100, $60 is 0), 15% practitioner sentiment from Signals and community threads, 15% operational fit (context window, open weights). A missing component drops out and the rest are reweighted. A model with no independent eval yet is provisional and ranks below every measured one. Within three points is a tie. Vendor-published figures are labelled and never counted. Prices are vendor list prices where published; blended figures assume three input tokens per output token.