Is deepseek-v3.1-terminus-medium good at Knowledge & Docs?
deepseek-v3.1-terminus-medium ranks #40 of 94 for Knowledge & Docs — usable. The top pick for this task is gpt-5.4-max.
deepseek-v3.1-terminus-medium on each Knowledge & Docs sub-task
| Help-Center Article | 100.0/100 | #22 |
| Grounded Docs (No Fabrication) | 99.0/100 | #56 |
| Right Doc Type | 77.5/100 | #30 |
| Procedure / Runbook | 42.0/100 | #75 |
Real examples, graded
WeakDocument only what exists (Ferrovia) — with an unsupported ask 44/100
“The model correctly addresses the user's request by confirming that bulk-approve and auto-approve are not supported, and it accurately describes the manual fallback process. However, it violates the strict grounding rule by inventing 'release notes on the Integrations page', which is not present in the provided facts. This fabrication drops the grounding score to 2 and caps the overall score.”
WeakNo data: write honestly (Tradewinds) 23/100
“The model completely failed the core instruction to not invent the feature or its steps. Instead of responding to the teammate to ask for the missing product facts, it wrote a generic, filler-heavy help article directed at an end-user that guesses at potential UI elements and menu paths (e.g., 'Shift History', 'Time Sheets', 'Settings'). This violates the strict grounding rules against fabricating UI.”
Frequently asked
Is deepseek-v3.1-terminus-medium good at Knowledge & Docs?
deepseek-v3.1-terminus-medium ranks #40 of 94 models we tested for Knowledge & Docs, scoring usable.
What is deepseek-v3.1-terminus-medium's strongest Knowledge & Docs skill?
Its best sub-task here is Help-Center Article.
This page is Spring Prompt, running
We just did this for every model. Do it for your prompt.
The rankings above come from running real tasks through real models and scoring every output. Spring Prompt is that same engine — pointed at your prompt, your test cases, and your definition of good.
- Generate test cases from your prompt — no eval set required to start.
- Compare models side by side with quality, cost and latency in one matrix.
- Optimise the winner until the scores say it's ready to ship.
Prompt × model results
12 test cases · 3 evals