MiniMax M3.1 Flash has lower provider-published standard API rates than Kimi K3.1 Pro for input, output, and cached input tokens. At a 3:1 mix of uncached input to output, the calculated cost is $0.2625 per million tokens for MiniMax and $4.00 for Kimi. OrcaRouter’s reported Plotly test gives a separate, single-run view of task costs.
Published API rates
The providers list these standard API rates in USD per million tokens:
| Billing category | Kimi K3.1 Pro | MiniMax M3.1 Flash |
| Regular input | $2.00 | $0.15 |
| Output | $10.00 | $0.60 |
| Cached input | $0.20 | $0.015 |
| Uncached 3:1 input-to-output blend | $4.00 | $0.2625 |
The Kimi rate card also lists cache writes at $2.00 per million tokens for a five-minute time-to-live (TTL) and $4.00 for a one-hour TTL. TTL is how long a cache entry is retained. Kimi says its rates exclude applicable taxes. MiniMax describes its API as pay-as-you-go, with charges based on actual usage; its pricing page lists the three rates above.
What the 3:1 mix costs
The blend assumes three input tokens for every output token, with no cache use. For Kimi, the calculation is (3 × $2.00 + $10.00) ÷ 4 = $4.00 per million tokens. For MiniMax, it is (3 × $0.15 + $0.60) ÷ 4 = $0.2625 per million tokens. That makes the Kimi blend about 15.2 times the MiniMax blend under this specific assumption.
These figures exclude cache hits, cache writes, tools, and applicable taxes. A real estimate needs the workload’s input and output totals, plus cached-token use. Kimi’s cache-write fees also depend on the selected TTL.
What OrcaRouter’s Plotly run found
On October 26, 2026, OrcaRouter reported one run per model on Plotly DataAnalyticsBench, a 43-question test. Kimi answered 42 questions correctly at a total cost of $13.47; MiniMax answered 41 correctly at $4.09.
That is a one-answer difference in this run: 42/43 is about 97.7%, while 41/43 is about 95.3%, a gap of roughly 2.3 percentage points. The single run does not establish how either model performs across repeated runs or other workloads.
Dividing the reported totals by 43 gives about $0.313 per Kimi question and $0.095 per MiniMax question, a ratio near 3.29 to 1. OrcaRouter’s rounded per-question figures, $0.31 and about $0.10, produce its stated 3.1-to-1 ratio. The task costs also reflect the tokens consumed in that run, not just the rate card.
Which figure belongs in your budget?
Use the published per-token rates to price a workload only after estimating its token mix. A task with more output, cached input, or cache writes will not match the uncached 3:1 example. For a Kimi estimate, count regular input, cached input, cache-write tokens by TTL, and output separately; then apply the corresponding rates. OrcaRouter’s one-run totals are useful as a task-specific example, not as a general bill forecast.
