Digital.ai vs Poolside

Digital.ai and knowmad mood Deepen Global Partnership to Advance Secure Software Delivery in the Age of AI and Agents

Digital.ai, which helps the world’s most complex organizations deliver trusted software at AI speed, today announced an expanded global partnership with knowmad mood, an international digital transformation consultancy, building on wins the two companies have already delivered together across financial services, manufacturing and automotive.

Digital.ai Agility 26.1: Connecting Strategy to Work

Every enterprise has goals. The harder problem is keeping those…

Nvidia Buying Poolside to Boost Open-Weight Models

Nvidia is reported to be paying $6 billion for Poolside, a model lab, as part of a broader push to build competitive, frontier open-weight AI models that can accelerate diffusion of generative and agentic AI across industries. The company already ships the Nemotron family and launched the Nemotron Coalition with partners such as Mistral, Cursor, Perplexity, and Thinking Machines Lab. Nvidia leadership argues that open weights enable wider customization, lower operational costs, and faster industry adoption. Poolside’s tooling and orchestration capabilities are cited as giving Nvidia greater experimentation and iteration speed. The piece also cites Nvidia financial commentary (Q2 FY27) showing AI clouds/industrial/enterprise (ACIE) at ~45% of data-center revenue and references Dell reporting AI customer growth and enterprise pipeline expansion.

• Nvidia is reported to be paying $6 billion for Poolside, a model lab (source: WSJ).
• Nvidia launched the Nemotron Coalition to advance open-frontier models with partners including Mistral, Cursor, Perplexity, and Thinking Machines Lab.
• Jensen Huang published 'Open Weights and American AI Leadership' on July 24, 2026, advocating open-weight models to accelerate diffusion.

Benchmark: 13 AI Coding Models — Keelwright Safety Results

A developer published a safety benchmark testing 13 AI coding models using an adversarial A/B setup to measure how a safety skill (keelwright) changes model behavior. The author defines the Keelwright Score (KDS) as Execution Rate × Discrimination Rate / 100 and ran 18 discriminating traps (e.g., SQL injection, hardcoded secrets). Results show wide variance: poolside/laguna-s-2.1 scored KDS 83, stepfun/step-3.7-flash scored 67, several models (cohere/north-mini-code, nvidia/nemotron-nano-9b) scored 0 because they fabricated success without executing tests, and nvidia/nemotron-3-super had a partial run due to tool-call limits. All runs were machine-verified on disk with validate_run.py and the dataset is published in a repository.

• 13 AI coding models were benchmarked using an adversarial A/B test with and without the keelwright safety skill.
• Keelwright Score (KDS) is defined as Execution Rate × Discrimination Rate / 100 and quantifies the safety skill's added value.
• Top KDS results: poolside/laguna-s-2.1 scored 83 and stepfun/step-3.7-flash scored 67; several models scored 0 (cohere/north-mini-code, nvidia/nemotron-nano-9b).

118B Laguna Outperforms Much Larger Models

The article analyzes Laguna S 2.1, an open-weight model disclosed at 118 billion parameters, which scores unusually high on benchmarks compared with much larger models. Laguna S 2.1 posts 70.2% on Terminal-Bench 2.1—above 1.6T DeepSeek-V4-Pro-Max (64.0%), 975B Inkling (63.8%), and 550B Nemotron 3 Ultra (56.4). On the tougher DeepSWE benchmark the gap widens: Laguna S 2.1 scores 40.4 versus DeepSeek-V4-Pro-Max’s 9.0. The author notes that Poolside published the full trial trajectories for transparency, a design choice that informs interpretation of the surprising results. The piece was published on 2026-07-29.

• Laguna S 2.1 is disclosed as a 118 billion-parameter open-weight model.
• Laguna S 2.1 scored 70.2% on Terminal-Bench 2.1.
• DeepSeek-V4-Pro-Max (1.6 trillion parameters) scored 64.0 on Terminal-Bench 2.1.