Step 5 Preview API Pricing
Step 5 Preview is StepFun's new flagship for agentic work, listed on September 20, 2026 with a 1M-token context and text, image and video input. StepFun's USD price list charges $1/M input, $2.70/M output and $0.05/M for a cache hit - 5% of input, where StepFun's Flash models charge 20%. Reasoning tokens bill as output. Checked by hand against platform.stepfun.ai.
Run the numbers.
Live calculator pre-loaded with Step 5 Preview's USD rates. The cache hit costs 5% of input, so the cache-hit slider moves this model's bill more than most; reasoning tokens bill at the output rate, and reasoning_effort (low, medium, high) is the lever on how many there are.
Real-world presets.
Repo-wide bug fix
Filing analysis
Screen-recording summary
Research planning turn
This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.
| Model | Input /M | Output /M | Effective blended | Context | Best for |
|---|---|---|---|---|---|
| Step 5 Preview Current | $1.00 cache $0.05 | $2.70 | $0.419 agentic 92/8 | 1M | StepFun's 1M-context agent flagship |
| Step 3.7 Flash | $0.20 cache $0.04 | $1.15 | $0.155 −2.7× | 256K | Cheaper StepFun multimodal agents |
| Step 3.5 Flash | $0.10 cache $0.02 | $0.30 | $0.0556 −7.5× | 256K | Cheap StepFun text reasoning |
| Kimi K3 | $3.00 cache $0.30 | $15.00 | $1.92 +4.6× | 1M | Chinese frontier coding alternative |
| GLM-5.3 | $1.40 cache $0.26 | $4.40 | $0.78 +86% | 1M | Zhipu's coding flagship |
| DeepSeek V4 Pro | $1.32 cache $0.044 | $3.96 | $0.569 +36% | 1M | Low-cost reasoning workloads |
| Gemini 3.8 Flash | $0.75 cache $0.075 | $3.75 | $0.481 +15% · promo | 1M | Mainstream multimodal work |
Frequently asked.
Practical Step 5 Preview pricing questions - which price list applies, what the cache does to the bill, and what 'preview' means here.
Q · 01 What is Step 5 Preview priced at? +
$1/M input on a cache miss, $0.05/M on a cache hit and $2.70/M output for step-5-preview. The cache-miss price includes writing new content to the cache, and the model's reasoning is billed as output alongside the final answer.Q · 02 Why does the Chinese price list show different numbers? +
Q · 03 How much does the cache save? +
$0.05/M, 5% of the input price, while StepFun's Flash models charge 20%. Under our standard blend (92% input tokens, 82% of them cache hits) the effective rate is $0.419/M.Q · 04 What does 'Preview' mean for production use? +
Q · 05 What are the limits and features? +
reasoning_effort at low, medium or high.Q · 06 How does it compare with Step 3.7 Flash? +
$0.20/M input and $1.15/M output with a 256K context. Step 5 Preview is five times the input price but only about 2.3 times the output price, with four times the context and a cheaper cache hit relative to its input rate.