AI this week: OpenAI DevDay’s GPT-6.1 Sol, Google’s Gemini 4 Argon, robots from an image lab
GPT-6.1 Sol nearly matches OpenAI’s flagship at a quarter of the cost, Google’s first Gemini 4 model targets knowledge work, and Black Forest Labs moves into robots.
October 9, 2026 · By Bohol AI
Mid-tier models caught up with flagships, Google finally refreshed its top model, and image-generation companies kept moving into robotics.
GPT-6.1 Sol
At its DevDay conference on September 29, OpenAI introduced GPT-6.1 Sol, which lands within a point of its flagship GPT-6 Astra on Artificial Analysis’ index at less than a quarter of the cost per task. It replaced GPT-6 Sol a week after that model’s debut, while a planned Astra upgrade was withdrawn days before launch.
Gemini 4 Argon
Google announced Gemini 4 Argon, its first Gemini 4 model, trained for long, complex problems in software engineering, law, finance and cybersecurity, with an emphasis on fewer made-up answers. For now only cybersecurity defenders can use it.
FLUX 3 Action
Black Forest Labs, known for its FLUX image generators, released FLUX 3 Action, a 7-billion-parameter model that turns camera feeds and text instructions into robot actions. It tops Nvidia’s RoboLab simulation benchmark, though it hasn’t yet been tested on real robots in the RoboArena benchmark.
Research: picking the right tools
Researchers in Shanghai and Hong Kong built HYSET, which chooses the best combination of tools for an agent’s task instead of judging each tool on its own.
What it means for Bohol
- Compare cost per finished task, not price per message. A slightly weaker model at a quarter of the cost is often the better deal.
- Models that make up fewer facts matter most in law, finance and health. Still check anything important against a real source.
Sources
- OpenAI: GPT-6.1 Sol
- Artificial Analysis: GPT-6.1 Sol
- Google: Gemini 4 Argon
- Black Forest Labs: FLUX 3 Action
- Nvidia: RoboLab leaderboard
- Hong et al., HYSET tool selection (arXiv)
- The Batch, issue 374 (DeepLearning.AI)