OpenAI cuts GPT-5.6 Sol pricing as agents move into shared workspaces

OpenAI has cut GPT-5.6 Sol API and credit pricing by more than 20% for three months, turning the latest frontier-model contest into a sharper cost-per-task fight. The wider GPT-5.6 family now combines programmatic tool calling, multi-agent execution and a 30-minute cache window, while OpenAI says Sol reaches 80 on its Coding Agent Index. The practical implication is immediate: teams can test longer-running agent workflows without treating every extra reasoning step as a budget failure. The temporary discount also gives developers a deadline to measure whether Sol’s claimed efficiency survives real workloads. Read OpenAI’s GPT-5.6 pricing update.

What else matters

OpenAI launches a Strategic Futures team

OpenAI has launched AI Futures, a blog and research effort focused on how transformative AI could alter power, institutions and individual agency. The opening essay explicitly says the work represents its author’s views, not company-wide policy, but it still marks a notable shift: frontier labs are now publishing arguments about governance architecture, not just model capabilities. Read the AI Futures launch post.

GitHub brings Copilot agents into Microsoft Teams

GitHub’s public preview lets teams start a Copilot cloud-agent session by mentioning @GitHub in a Teams channel, thread or direct message. Everyone can add context and steer the work; repository writers can trigger changes. The integration also adds an extra approval option for agent-authored pull requests, making human oversight part of the workflow rather than an afterthought. Read GitHub’s Teams integration announcement.

DeepMind turns EVE Online into an agent research sandbox

Google DeepMind says its new partnership with Fenris Creations will use EVE’s persistent, player-driven worlds to study continual learning, memory, long-horizon planning and multi-agent behaviour. The programme starts with an offline instance, then moves toward controlled experiments in EVE Frontier. That sequencing matters: the lab is treating open-ended environments as a safety and evaluation problem before a product surface. Read DeepMind’s games research update.

The common thread: agents need operating environments

Across these announcements, the competitive unit is no longer a model leaderboard alone. OpenAI is packaging parallel agents and tool execution; GitHub is embedding an agent in team discussion; DeepMind is building persistent worlds in which agents must remember, plan and negotiate. The next meaningful comparisons will be end-to-end completion, oversight and operating cost—not isolated benchmark scores.