Salesforce Koa
Enterprise CRM reasoning model for Agentforce built on NVIDIA Nemotron
Context window
1000000
Input / 1M tokens
Not announced
Output / 1M tokens
Not announced
Provider
Salesforce
Data verified 2026-09-19
Salesforce Koa is a CRM reasoning model purpose-built to help agents reason through complex, multistep workflows and use tools to complete enterprise work. Built by post-training NVIDIA Nemotron 3 Super with a proprietary synthetic dataset derived from nearly 27 years of Salesforce CRM deployments and enterprise knowledge across 14 industries. Trained entirely on public and synthetically generated data with no customer data used.
Capability index
Relative estimates (0-100) to place this model against its peers, grounded in published benchmarks.
How to access it
Currently in pilot with select customers. General availability expected winter 2026 in the US through Salesforce Agentforce platform.
Strengths
- ✓Specialized for CRM tasks like updating opportunities, routing cases, and scheduling follow-ups
- ✓Achieves 3x fewer errors than leading models on Salesforce's CRM benchmark
- ✓Operates on open-weight model with full Salesforce control over weights and inference
- ✓Enterprise knowledge grounded in 27 years of CRM deployment patterns
Best for developers who...
When to choose it (and when not to)
Reach for Salesforce Koa when...
- →When you need a specialized CRM reasoning model vs general-purpose LLM
- →For organizations prioritizing data control and private deployment
- →When CRM-specific performance on multi-step workflows is critical
Look elsewhere if...
- ✕For general-purpose language tasks outside CRM domain
- ✕If maximum frontier model performance on general benchmarks is needed
- ✕For organizations not using Salesforce Agentforce platform
How to use it
Quickstart
Not announcedNot announcedKoa is accessed through Salesforce Agentforce platform, not via standard API
API model id: Not announced
Benchmarks
| Benchmark | Score | Notes |
|---|---|---|
| Salesforce CRM Benchmark | Matches or exceeds leading models with 3x fewer errors | Includes tasks like updating opportunities, routing cases, scheduling follow-ups |
| Salesforce CRM Bench | 3x fewer errors; 11% more precise at action calls; 2.1x better context recall; 15% better long conversation context retention | Measures real-world CRM tasks: updating opportunities, routing cases, scheduling follow-ups |
Source: Salesforce Official Announcement
Compare Salesforce Koa
Compare Salesforce Koa with any other model
Build a comparison →All model comparisons →