AI model
NVIDIA's family of open large language models, built for enterprise AI, reasoning, and agentic applications across various model sizes.
NatWest and Nvidia cut misheard words 20% in the Spending Insights voice tool, now open to 75% of Royal Bank of Scotland customers. The announcement did not disclose a starting error rate, so the gain cannot be sized in absolute terms.
Reality
Honeywell's CTO says Forge Cognition will run with people in the loop for probably two to three years, autonomous mode off. The evidence so far is one vendor's pace, tested in Nvidia's own headquarters; how fast buyers move is still a forecast.
Reality
Palantir CTO Shyam Sankar said on Face the Nation that the Pentagon replaced Anthropic's AI models with no operational impact. No cost or quality figures were reported, so operators weighing a model-agnostic design still have to test a swap on their own work.
Reality
IBM's Bob coding agent now runs air-gapped inside customer networks on two supported local models, Nvidia's Nemotron and Poolside's Laguna. Banks can keep their code in-house, though fully local work runs on a shorter model list than the Claude, Mistral and Granite mix IBM pitches for Bob.
Reality
Capsule Security has shipped an evaluator that runs inside an agent's execution path and can stop an intended action before it executes. Doing that requires adding a chokepoint to the agent's execution path.
Reality
Nvidia has named eight Australian operators and a 2027 target of up to two gigawatts. The announcement leaves out per-site dates, price, customers and power supply, which is the part a buyer has to sign.
Reality
NVIDIA says its critical material allocation is reworked by hand every week and that the Vera Rubin supply chain is twice the size of Grace Blackwell's, evidence that supports codifying planner judgment though the text available stops short of showing it worked.
Perspective Coverage
4 publishersReality
Salesforce and Nvidia post-trained Koa on simulated CRM workflows and claim three times fewer errors than leading general-purpose models. The published record describes the training data two different ways.
Perspective Coverage
3 publishersReality
The workflow gives each subagent one job and its own acceptance test, repairs what it classes as safe, and sends an uncertain semantic label back to a human with a proposed next step.
Reality
Cisco's Splunk unit is repositioning around telemetry it expects agents to generate faster than people can read, with a new fabric that queries Snowflake, Databricks and object stores in place. Pricing was not disclosed.
Reality
NVIDIA's tutorial post-trains a 4B model into a Franka manipulation policy that runs on Jetson Thor with about 0.6 seconds of latency slack per cycle. Closed-loop success is 22.9%.
Reality