Skill · Business Ops
Run It on DeepSeek, Qwen, GLM or Kimi
Moves a prompt, skill or API call to DeepSeek, Qwen, GLM or Kimi with the right model, effort and parameters, then checks three real inputs side by side before you switch.
When to use
- A recurring job (summaries, extraction, drafting, classification) costs too much on a frontier model and you want a cheaper one that still passes.
- You need a prompt or skill to run on open weights you host yourself.
- An API call that worked on another model errors or behaves oddly on DeepSeek, Qwen, GLM or Kimi.
- You need to choose among the four for a job with images, very long documents or tool calls.
When not to use
- The job is high-stakes and the current model already meets the bar; switching adds risk unless cost forces it.
- You want a general ranking of models. This skill does not rank; the side-by-side check tells you whether the switch works for your job.
- The inputs are confidential and you have not checked the hosted provider's data terms. Self-host, or check first.
Example
User (illustrative): "Move our weekly support-ticket summary from GPT-6.1 Sol to something cheaper. API, one call a week, about 80k tokens of tickets, text only. The request sets sampling parameters. Tickets contain customer emails."
Result (excerpt)
Switch plan: GPT-6.1 Sol → DeepSeek V4.1-Flash (fallback: GLM-5.3-Flash)
Why: text-only summaries are a quick task; 80k tokens fits the 1M context easily.
Settings
- Model: deepseek-flash, DeepSeek's OpenAI-compatible endpoint (change base URL, key, model id)
- Depth: reasoning_effort low
- Remove: all sampling parameters (no effect in thinking mode)
Gotchas for this setup
- Customer emails go to DeepSeek's API: check its data terms first, or run the open weights yourself.
- If you later add tool calls, send back every reasoning_content.
Check before switching
| Input | Criterion | Current | Target |
|---|---|---|---|
| Week −1 tickets | Every escalation listed | pass | ? |
| Week −2 tickets | Category counts match the export | pass | ? |
| Week −3 tickets | No invented customer names | pass | ? |
Switch if: the target passes all nine checks.
Illustrative example written by Meevra to show the format; actual output will differ.
What is inside
The full skill has these sections. The ones marked “in the pack” come with the download.
- When to use on this page
- When not to use on this page
- Inputs to ask for in the pack
- Workflow in the pack
- Rules in the pack
- Output format in the pack
- Example on this page
- Reference: model facts (Oct 2026) in the pack
- Model notes (Oct 2026) in the pack
- Troubleshooting in the pack
- Paste-in version (for chats without skill support) in the pack
Get Run It on DeepSeek, Qwen, GLM or Kimi
The full skill is a SKILL.md folder you install once and reuse: the workflow, the checklist and the answer format written out, with notes for each model. It comes in the Business Ops pack (11 prompts · 7 skills · 2 free), with a QUICKSTART, setup notes in every item and 12 months of updates.
All-Access 39 € vs 7 packs × 12 € = 84 € · less than buying 4 or more packs one by one
Stuck? Email support@meevra.com — we'll help you set it up.