✸
Support reply drafter
SupportDraft an accurate, empathetic support reply from the ticket + a KB snippet — and flag when it should be escalated to a human instead of guessing.
The routing recipe
Job typeGeneral chat / assistantPolicyFastest that passes
routes to
GPT-OSS 20B groq
groq:openai/gpt-oss-20b
90.4
match score
77.7
task fit
100
cost
96.4
speed
Why this model from benchmarks & capabilities
- "General chat / assistant" leans hardest on instruction; GPT-OSS 20B scores 78/100 there — #23 of 40.
- Strength on this class of work shows up on LMArena Elo, MT-Bench, IFEval, where OpenAI (open-weight) models are competitive.
- Cost: free tier. Speed: ~1000 tok/s on Groq LPU.
- Chosen over GPT-OSS 120B for cheaper & faster.
| Capability the job needs | Weight | GPT-OSS 20B | Catalog rank |
|---|---|---|---|
| instruction | 0.80 | #23 of 40 | |
| knowledge | 0.60 | #26 of 40 | |
| speed | 0.60 | #1 of 40 |
Relevant benchmarksLMArena Elo MT-Bench IFEval
EconomicsCost: free tier.
Speed: ~1000 tok/s on Groq LPU. vs runner-upChosen over GPT-OSS 120B for cheaper & faster. (GPT-OSS 120B, score 85.2)
Speed: ~1000 tok/s on Groq LPU. vs runner-upChosen over GPT-OSS 120B for cheaper & faster. (GPT-OSS 120B, score 85.2)
Example result
Sample input ▾
Ticket: "Exports have failed all morning with a 500 — this is blocking our board deck due at noon." KB snippet: "Large CSV exports over ~50k rows can time out; use Filtered Export, or the API with pagination."
Hi — sorry this is hitting you right before a deadline, let's get you unblocked.
The 500 is almost certainly an export timeout: exports above ~50,000 rows can exceed the limit. Two fast options:
1. Use Filtered Export (date or segment) to pull the deck data in smaller chunks.
2. If you need everything, the API paginates and won't time out — happy to share a ready-to-run snippet.
Want me to generate the filtered export for your board metrics right now so you have it before noon?
Test it on your own data