Alibaba’s Qwen4-Max scores 43 on Artificial Analysis’s Intelligence Index, against 29 for StepFun’s Step-5 Preview. But Step-5 costs less per index task and offers downloadable weights, making this a useful comparison of trade-offs rather than a tidy leaderboard victory.
Watch Desk analysis
What happened
OrcaRouter compares the two models using the same index. It lists Qwen4-Max at $3.27 per index task and Step-5 Preview at $2.35. On cost per index point, however, its figures are closer: 7.6 cents for Qwen4-Max and 8.1 cents for Step-5. These are the comparison’s reported measurements, not a universal price for every workload.
The practical differences extend beyond the scores. OrcaRouter says Step-5 Preview has downloadable weights and an API price of $1 per million input tokens and $2.70 per million output tokens. Qwen4-Max has no published weights, and its listed API rates are $1.50 input and $4.50 output per million tokens in Singapore. Step-5’s output speed is listed at 124 tokens per second; OrcaRouter gives no measured speed for Qwen4-Max.
Why it matters
The comparison puts a sharper shape on a familiar choice: higher measured capability, or lower costs and the option to run the model yourself. For teams that need to inspect, adapt or host weights, Step-5 has a concrete advantage. For teams choosing by index score, Qwen4-Max leads, but that lead does not settle which model will perform better on their own work.
OrcaRouter also notes that Step-5’s benchmark table was published by its maker, StepFun. That makes the scores useful for orientation, not a neutral final verdict. A leaderboard is a starting point; your workload is the rather less glamorous referee.
Our read
This is worth reading as a buying guide, not a declaration of one model’s overall superiority. Compare representative tasks, include the actual API or hosting costs, and check the latency you can tolerate. If downloadable weights matter, that may decide the question before the index score gets a vote.
What to watch
- Whether independent evaluations compare both models on the same tasks and settings.
- Whether Qwen4-Max receives published latency measurements or downloadable weights.
- Whether StepFun clarifies Step-5’s preview status and output limits.
Discussion spark: Would you choose Qwen4-Max’s higher index score, or Step-5’s lower listed costs and downloadable weights for a production system?
Sources and evidence
- Qwen4-Max vs Step-5 Preview: cost per index point compared – OrcaRouter (16 November 2026, 10:09 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.