StepFun’s Step 5 Preview is available with downloadable weights, published pricing and text, image and video input; DeepSeek’s V4.1 Pro is not a released model, according to OrcaRouter. Its comparison instead puts Step 5 against DeepSeek V4 Pro, the available text-only model, giving teams a more useful choice than comparing a product with a name on paper.
DeepSeek Watch analysis
What happened
OrcaRouter says DeepSeek’s only public mention of V4.1 Pro was a proposed API change that would have redirected requests to V4.1 Flash until the Pro model launched. DeepSeek withdrew that plan after developers objected, and has not published a release for V4.1 Pro, the comparison says. The article therefore compares Step 5 Preview with V4 Pro, not the unreleased V4.1 Pro.
The practical differences are substantial. OrcaRouter lists Step 5 Preview as a 600-billion-parameter model with 27 billion active parameters, multimodal input and downloadable weights in three precisions. DeepSeek V4 Pro is listed as a 1.6-trillion-parameter, text-only model, with a 384,000-token maximum output and lower off-peak prices. OrcaRouter’s full comparison also details pricing, context limits and benchmark information.
Why it matters
The comparison makes two different buying decisions easier to separate. For text-heavy work with repeated prompts, OrcaRouter says DeepSeek V4 Pro’s lower cache-hit price can be a meaningful advantage. For jobs involving images or video, Step 5 Preview supports inputs that V4 Pro does not. Step 5’s weights are available to download; OrcaRouter says it is not available through its own model router.
There is a useful wrinkle in StepFun’s documentation: OrcaRouter notes that one source gives a 64,000-token output limit while another says “up to 1M”. It advises treating 64,000 as the hosted API limit until a longer output is demonstrated. Benchmarks and price comparisons are also only as useful as the configurations and workloads behind them, so this is a guide to trade-offs, not a universal winner’s podium.
Our read
The headline is not that one model beats the other. It is that Step 5 Preview is a real, multimodal option, while V4.1 Pro is not yet something developers can call. If you are choosing between the available models, match the model to your inputs and test the actual workload; do not build around an unreleased name. A model identifier is not a deployment plan, however confidently it sits in a spreadsheet.
What to watch
- Whether DeepSeek announces or releases V4.1 Pro, with an endpoint and pricing.
- Whether StepFun clarifies the discrepancy over Step 5’s maximum output length.
- How Step 5 performs on independent tests using workloads comparable to those used for V4 Pro.
Discussion spark: When choosing a model, would you prioritise lower costs for repeated text prompts, or pay more for multimodal input and weights you can download?
Sources and evidence
- Step 5 Preview vs DeepSeek V4.1 Pro: Mana yang Benar-Benar Ada – OrcaRouter (16 October 2026, 07:05 UTC)
not affiliated with or endorsed by DeepSeek