Salesforce and Nvidia’s new reasoning model is everything the AI labs should fear
- ID
- 24596
- Status
- summarized
- Published
- 15 Sep 2026, 8:00 PM
- Fetched
- 15 Sep 2026, 9:26 PM
- Provider
- TechCrunch
- Category
- technology
- Original URL
- https://techcrunch.com/2026/09/15/salesforce-and-nvidias-new-reasoning-model-is-everything-the-ai-labs-should-fear/
- Source URL
- https://techcrunch.com/feed/
Summary
- Score
- 6.5
- Created
- 15 Sep 2026, 9:28 PM
- Tags
- Audience
- developersai_agent_userssaas_founders
What happened
Salesforce and Nvidia unveiled Koa, Salesforce's first reasoning model, built on Nvidia's open-weight Nemotron and post-trained for sales, marketing, and customer-support tasks. Koa will be offered as an alternative to closed frontier models like Claude and ChatGPT within Salesforce's Agentforce platform, promising lower token usage, no customer data ingestion, and automatic routing through an AI gateway.
Why it matters
If you build AI agents or SaaS products, this signals a viable pattern: take an open-weight model, post-train it for a narrow vertical, and route to it instead of always calling expensive frontier APIs. Founders should evaluate whether fine-tuning an open-weight model for their domain could cut inference costs and data-leakage risk versus defaulting to OpenAI or Anthropic for every reasoning step.
Discussion angle
Is the 'open-weight + post-train for a vertical + gateway routing' approach practical for a small Malaysian startup team, or is it only viable at Salesforce scale with Nvidia partnership?