Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- OpenCode lists Step 5 Preview Free. Official posts say one free week, without an exact ending clock or time zone.
- StepFun, OpenRouter and Zen use different IDs; do not transfer the free model name across platforms.
- Standard uncached input/output are $1/$2.70 per million tokens. StepFun cached input is $0.10, while OpenRouter lists $0.05; check each channel.
- 1M is context and maximum output is 64K. Zen's provider-specific no-retention/no-training statement must stay tied to the actual channel.
Find the version labeled Free
StepFun announced Step 5 Preview on OpenRouter October 8, 2026, with one-week free access rolling out through OpenCode, Cline and other tools. OpenCode also announced a free week, and current Zen docs list Step 5 Preview Free.
Zen lists input, output and cache reads as Free for a limited time. Neither posts nor docs give an exact ending time, time zone or per-account allowance; do not calculate a universal expiry from post time.
Check the actual tool model list rather than Step 5 alone. Other clients retain their own rollout and rules; rolling out does not establish simultaneous access in every version, region or account.
Keep three platforms' IDs separate
StepFun API uses step-5-preview; OpenRouter uses stepfun/step-5-preview. Zen's free ID is step-5-preview-free, configured in OpenCode as opencode/step-5-preview-free.
First sign into the official Zen page, follow its onboarding instructions for billing details and obtain your own API key. Keep that service credential in your own client configuration.
Second run /connect in OpenCode's terminal UI, select OpenCode Zen and enter the key.
Third open /models and select Step 5 Preview Free. The configuration name is opencode/step-5-preview-free.
Generic onboarding does not explicitly waive payment details for this free model or promise unlimited requests. Model price, account conditions and limits are separate: confirm ID and Free status, then your account prompts.
Standard APIs retain usage pricing
As of October 10, StepFun Standard API rates per million tokens are $1 uncached input, $0.10 cached input and $2.70 output. Use your chosen channel's current table rather than the model name alone.
OpenRouter also lists $1 uncached input and $2.70 output, but cached reads remain $0.05. Do not call the rates identical. Its full catalog did not list a Step 5 free variant; Zen's Free entry is separate.
Tokens meter the processed information. StepFun bills reasoning and final answers as output, so visible-answer length can understate charges. Writing new content to cache is included in uncached-input pricing.
A hypothetical 100K uncached-input/10K billed-output Standard request costs about $0.127. No call was executed, and tools, retries and other services are excluded.
Standard pay-as-you-go and Step Plan subscriptions have different usage rules. Personal Standard API accounts require identity verification and rate tiers based on cumulative cash top-ups. OpenCode's free week is not a StepFun recharge benefit.
A 1M window is not 1M output
Docs list 1M context and maximum input, but maximum output is 64K. Large input capacity does not mean equally long answers.
The model natively accepts text, images and video and returns text, supporting tools and structured output. A particular client's media or parameters depend on its integration; model support does not prove every editor implemented them.
Search, code execution and tools are supplied by the application. The model does not directly access your computer or external services. For hypothetical debugging, the app must expose appropriate files/tools; changing models creates neither connections nor approval.
Zero retention applies to the named channel
Zen specifically states that the provider for Step 5 Preview Free retains no data and does not train on it. That is not a promise for all StepFun products, OpenRouter routes or local client logs.
Zen documents exceptions for other models, so its brand alone does not establish uniform privacy. Check actual provider, client sharing/logging and files being sent after selecting the model.
Use a small public project to evaluate limited free access. Before private code, customer information or keys, check team authorization and data rules. Free access and zero retention do not make all material suitable for every channel.
Check selection and billing before use
Confirm the Free Zen model or a Standard API model whose charges you accept. Review account and fees, retain current free status and avoid unknowingly defaulting to paid fallback models.
If also using paid Zen models, check auto-reload and monthly usage limits separately. Generic auto-reload adds $20 below $5 balance and can be disabled. A $20 monthly usage limit is not a $20 payment cap: reloads and fees may cause higher charges.
Evaluate a small task's answers, tools and actual usage before expanding. The free window, long context and official capabilities justify evaluation, not automatic replacement of working models or acceptance of every task.
Sources
Frequently Asked Questions
Why can other models appear while the main model is free?
OpenCode may use low-cost Haiku, Nano or Flash for session titles. Free main-model pricing does not make every auxiliary request free. Check actual request/model records rather than assuming Step 5 Free charged.
If Free is missing, can I select ordinary Step 5 for free?
No. Standard StepFun/OpenRouter APIs have usage rates. Verify your client, channel and model list; changing a suffix does not grant promotion eligibility.
Does repeating input guarantee cache pricing?
Only actual cache-hit input uses the cached-read rate. Other input remains uncached-priced. Use returned usage/cache information; repetition alone does not guarantee hits, and cache rates do not apply to output.