Nvidia and Reflection Make Open Models Enterprise-Ready

 | 

Highlights

Nvidia and Reflection Build the American Open Stack

• Nvidia released Nemotron 3.5 Super VL on Oct. 15, an open 120-billion-parameter model that reads documents, charts and web pages. Parameters are the internal settings a model learns in training. Open weights means a company can download the model and run it on its own servers. Nvidia said it runs five times faster than comparable open models. Spend management FinTech Ramp is tuning it for financial document analysis, and Scale AI trained a version for a client’s merger due diligence work. Nvidia said a bank could use it to review credit files against its own lending rules without sending data to the cloud.
• Reflection AI released the full weights for Beam on Friday (Oct. 16) under a free commercial license. The Nvidia-backed startup first previewed the 501-billion-parameter model two weeks ago as a U.S. answer to Chinese open models. It launched with more than 40 partners, including AWS, Google Cloud, Microsoft and Oracle. Dell and HPE will deliver it for on-premises and air-gapped environments.
• The Institute of Foundation Models and Inception42 released Jais 3 on Oct. 15, a family of six open Arabic models from 900 million to 375 billion parameters. The Abu Dhabi-based group also released Jais Speech for Arabic speech recognition. The models handle regional dialects and are free to download.

DeepSeek and MiniMax Cut Prices Again

• DeepSeek updated its low-cost Flash model on Saturday (Oct. 17). The Chinese lab said the new version scores higher on every benchmark it reported, finishes tasks faster and writes shorter answers, which lowers the cost per job. The weights are free to download under an MIT license, the most permissive kind.
• MiniMax launched M3.1 Flash on Oct. 19. The Shanghai-based lab said the model runs nearly three times faster than its predecessor and cuts usage costs by 70%. It scored 76.2 on BankerToolBench, a test of multi-step financial tasks and compliance work. Pricing is $0.15 per million input tokens and $0.60 per million output tokens. Tokens are the small chunks of text AI companies use to meter usage. Weights will be released later.

Inception and Perplexity Chase Speed

• Inception Labs introduced Mercury 3 on Monday (Oct. 19). The startup builds models on a different technical approach, called diffusion, that generates text in parallel instead of one word at a time. Inception said Mercury 3 matches or beats OpenAI’s GPT-6 Luna and Anthropic’s Claude Haiku 5.5 while running four to eight times faster. Lorikeet, which builds support agents for regulated businesses, told Inception the model beat rivals on both speed and accuracy in early testing.
• Perplexity released two open decision models on Friday. Decision models do not write text. They pick from a fixed set of options and return a probability. The new versions answer several questions about the same input in one pass, which suits tasks like routing tickets or screening transactions.

The open model market is no longer a Chinese monopoly. For banks and insurers that could not use Chinese models for security reasons, the choice of what to run in-house just got wider.