Tech • AI • Robotics • Game

VIDEO
ENFR
TodayPlay for freeBloxLeaderboardShortsTop StoriesFor youTopicsVideosYT channelsArchivesSearchFavorites

OpenAI DevDay 2026: GPT-6.1 Sol, Dots, ChatGPT Space, White House accord

AIThursday, October 1, 2026· 20 videos

Briefing

Audio player
0:00 / 0:00

OpenAI unveils GPT-6.1 Sol

OpenAI positioned GPT-6.1 Sol as its new value frontier model, priced at $2 per million input tokens and $10 per million output tokens, versus GPT-6 Astra at $10 and $50. The company said Sol comes close to Astra on coding, computer use and professional workflows while dramatically lowering operating cost. On cited tests, it reached 75% on Deep Sway, 31.7% on Automation Bench, and 57% on Terminal Bench Science at $5.47 per task. The broader message is clear: frontier-adjacent reasoning is being pushed into mid-tier pricing.

Dots push ChatGPT into agents

The centerpiece of OpenAI's product slate was Dots, persistent cloud-based agents designed to keep working without constant prompting. Each Dot gets its own browser and cloud computer, can connect to more than 4,000 apps, and operates across ChatGPT, web, desktop, mobile, Slack and Microsoft Teams. Unlike a standard assistant, a Dot can monitor conditions, trigger tasks, maintain memory and return only when approvals or decisions are needed. The move reframes AI from reactive chat into a semi-autonomous digital co-worker.

ChatGPT Space becomes collaborative hub

ChatGPT Space extends that strategy by turning ChatGPT into a shared workspace with pages, subpages, dashboards, files and embedded AI help. Access is rolling out to Pro, Business and Enterprise users across web and desktop, with mobile following. The interface separates ongoing projects from ordinary chat, letting teams pin conversations, organize assets and run work in persistent contexts. It signals OpenAI's ambition to own not just the model layer, but the place where office work happens.

OpenAI sells cheaper models, premium compute

Behind the launches sits a sharper commercial model: lower token prices overall, but a premium for scarce compute and always-on execution inside OpenAI's cloud. The pricing ladder now spans GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, with Luna roughly 100 times cheaper than Astra for lightweight tasks. New products such as the Decision API package that low-cost inference for bounded operational choices like classification or confidence-scored outputs. The strategy suggests OpenAI wants customers to buy both intelligence and the workspace that contains it.

Gemini 4 Argon enters limited release

Google DeepMind has begun limited access to Gemini 4 Argon through its Fairwind program, initially for selected cybersecurity partners and trusted testers. Introductory pricing matches GPT-6.1 Sol at $2 per million input tokens and $10 per million output tokens, before a planned increase to $4 and $20. Its standout specification is a jump from 64,000 to 1 million output tokens, aimed at long-running software and agentic workloads. Reported benchmark results include 77.9% on SWE-bench 1.1, positioning Argon as a serious rival in long-horizon multimodal work.

White House signs superintelligence accord

A White House dinner produced a new Accord on Super Intelligence signed by leaders from OpenAI, Anthropic, Google, Meta, xAI, Nvidia and others. The four commitments cover internal capability and alignment controls, empowered internal verification teams, external audits, and independent board-level oversight. The scope explicitly includes cybersecurity, biosecurity and chemical threats, reflecting Washington's focus on frontier model risk. The event also served as a show of political access as AI policy hardens into industrial policy.

Open-source split sharpens in Washington

The biggest unresolved policy fault line was whether advanced open-weight models could face restrictions within the next 6 to 12 months. Safety-focused companies, especially Anthropic, argue the cyber gap between top proprietary systems and open models is narrowing fast enough to justify intervention. Discussion reportedly centered on possible controls at the inference, data-center, or chip level rather than a simple publication ban. The result is a growing split between firms advocating openness and those seeking tighter safety guardrails.

Meta Mus raises consumer agent stakes

Meta's new agent Mus, launched in the United States and Canada on September 8, is quickly emerging as a mass-market counterpoint to enterprise-focused agent launches. It connects to email, calendars and bank accounts to book services, manage subscriptions, place calls and automate personal admin, reaching the top of the free App Store in the U.S. on September 18. Pricing is described in a range of roughly $20 to $100 per month, with the most powerful features requiring access to highly sensitive personal data. Its rapid uptake sharpens the privacy and consumer-protection debate around AI agents acting directly in the real world.

Videos covered

Previous briefings · AI