Kimi K3.1 Pro vs Claude Sonnet 5.5

Verdict

Claude Sonnet 5.5 scores higher — 55 at max vs 46 at its only setting. Kimi K3.1 Pro costs 63% less per task ($3.68 vs $10.02). Per index point: $0.1822 for Claude Sonnet 5.5 vs $0.0800 for Kimi K3.1 Pro.

Kimi K3.1 Pro ranks #25 of 26 on cost per task among open-weights models larger than 150B parameters that AA prices; the class holds 129 models. Claude Sonnet 5.5 ranks #66 of 73 on cost per task among closed-source models above $1 per million tokens (blended 3:1) that AA prices; the class holds 237 models.

Our pick: Claude Sonnet 5.5. The price sentences above use it as the baseline; scope: API access.

• Kimi K3.1 Pro — Take Kimi K3.1 Pro when the bill binds: 46 vs 55 index points, $3.68 vs $10.02 per task, 37.9 vs 32.6 minutes — it gives up 9 points and saves $6.34 per task but is 5.3 minutes slower.
• Claude Sonnet 5.5 — Take Claude Sonnet 5.5 when score or latency binds: 55 vs 46 index points, $10.02 vs $3.68 per task ($0.70 per extra point), 32.6 vs 37.9 minutes.

Side by side

Where each wins

Kimi K3.1 Pro

• Cost per task (USD): $3.68 vs $10.02
• Cost per index point: $0.0800 vs $0.1822
• Level on Input modalities (text, image)
• Level on Input $/M tokens ($2.00)
• Level on Output $/M tokens ($10.00)
• Level on Context window (1M)

Claude Sonnet 5.5

• Intelligence Index: 55 vs 46
• Minutes per task: 32.6 min vs 37.9 min
• Cache discount: 95% vs 90%

Sources and conflicts

• OpenRouter lists moonshotai/kimi-k3.1-pro (exact match). Input price: AA $2 per million tokens vs OpenRouter $0.2375 — an 8.4× gap. This page uses AA's figure throughout and does not average the two. Modalities: AA records text and image; OpenRouter's route declares image, text, video.
• OpenRouter lists anthropic/claude-sonnet-5.5 (exact match). Input price: AA $2 per million tokens vs OpenRouter $2. This page uses AA's figure throughout and does not average the two.
• Prices and input modalities come from Artificial Analysis only; the second source is used to cross-check, never to union or average.
• A differing modality string is not a conflict: AA records what the model accepts, a provider records what its own route declares. Different objects, both reported here.

Data

Data as of 2026-11-12 · Artificial Analysis Intelligence Index v5.0.1, highest reasoning effort per model. Prices from the same pages.
Every cell is tied to a verbatim excerpt of that model's own Artificial Analysis page.
N/A (AA: Unknown) = AA publishes no value in that cell. It is not 0, and it is not 'unreleased'.
'not applicable (AA prints `-`)' = AA lists no per-token price for this model, so the cell does not exist. It is not the same as N/A and not the same as a withheld figure.
'not computable (…)' = the cell is derived from two others and one of them has no figure. The bracket names which one — it is not a value, and it is not the same as N/A.
'not printed on its AA page' = the page lists prices but withholds that figure; it is not treated as equal to anything.
'unverified' = we have not read that cell on that page. It is not a value and not a tie.
Token rows are AA's Verbosity block: input and output totals for the whole Intelligence Index run. 'Output/input ratio' is AA's own label and means output / input — it is not output's share of the total.
AA publishes no measured cache-hit rate for either model (it prints a fixed 7:2:1 cache-hit/input/output blend instead), so no cell records one.