❏
Call summary → CRM
SalesCondense a call transcript into a CRM-ready note: summary, action items with owners, sentiment, and the fields to update.
The routing recipe
Job typeSummarizationPolicyCheapest that passesConstraintsctx ≥ 32k
routes to
Llama 4 Maverick 17B-128E Instruct nvidia
nvidia:meta/llama-4-maverick-17b-128e-instruct
90.6
match score
86.1
task fit
100
cost
53.4
speed
Why this model from benchmarks & capabilities
- "Summarization" leans hardest on instruction; Llama 4 Maverick 17B-128E Instruct scores 94/100 there — top-2 of 32 chat models.
- Strength on this class of work shows up on LiveBench (language), grounding & summarization suite, where Meta models are competitive.
- Holds the 16k+ context this job needs (this model: 1M).
- Cost: runs on NVIDIA free credits / self-hostable NIM — effectively $0 marginal. Speed: ~45 tok/s (estimated).
- Chosen over Qwen3 Coder 480B-A35B Instruct for higher task-fit (86.1 vs 82.6) & more context.
| Capability the job needs | Weight | Llama 4 Maverick 17B-128E Instruct | Catalog rank |
|---|---|---|---|
| instruction | 0.90 | #2 of 32 | |
| long_context | 0.90 | #5 of 32 | |
| speed | 0.50 | #30 of 32 |
Relevant benchmarksLiveBench (language) grounding & summarization suite
Why it qualifiesHolds the 16k+ context this job needs (this model: 1M).
EconomicsCost: runs on NVIDIA free credits / self-hostable NIM — effectively $0 marginal.
Speed: ~45 tok/s (estimated). vs runner-upChosen over Qwen3 Coder 480B-A35B Instruct for higher task-fit (86.1 vs 82.6) & more context. (Qwen3 Coder 480B-A35B Instruct, score 90.4)
Speed: ~45 tok/s (estimated). vs runner-upChosen over Qwen3 Coder 480B-A35B Instruct for higher task-fit (86.1 vs 82.6) & more context. (Qwen3 Coder 480B-A35B Instruct, score 90.4)
Example result
Sample input ▾
30-min discovery with a 120-person healthcare SaaS; they need HIPAA + SOC 2, evaluating us vs Vanta, decision in 6 weeks, blocked on getting their CISO on a call.
Summary:
• 120-person healthcare SaaS; needs HIPAA + SOC 2 in parallel.
• Competitive eval vs Vanta; differentiator = our HIPAA evidence automation.
• Decision in ~6 weeks; gated on a CISO call.
Action items:
• Rep — send HIPAA + SOC 2 combined POC plan — Fri
• Champion — get CISO on a 30-min call — next week
Sentiment: 4/5 (engaged, time-bound).
CRM: Stage → Evaluation · Next step → CISO call · Close → ~6 weeks.
Test it on your own data