- NATURAL 20
- Posts
- OpenAI DevDay 2026 Brings Dots, GPT-6.1 Sol and a New Agent Platform
OpenAI DevDay 2026 Brings Dots, GPT-6.1 Sol and a New Agent Platform
PLUS: Claude Sonnet 5.5 gets faster, Nvidia moves agent safety into silicon, and xAI launches Team Bots.
Wes Roth
September 30, 2026
In partnership with
Chemical-Free Farming Is Now Open To Investors
For decades, farmers have relied on herbicides and heavy equipment to control weeds.
Greenfield Robotics is working to change that.
BOTONY robots navigate crop rows and mechanically remove weeds, giving farmers an alternative to blanket herbicide applications and intensive tillage.
The potential benefits are significant:
Clean food supply
Reduce reliance on agricultural chemicals
Reduce unnecessary soil disturbance
Help farmers protect crops from weed competition
Support more sustainable farming practices
And the market is enormous.
Greenfield estimates more than 250 million acres of broadacre cropland in the U.S., with more than 100 million acres in its initial serviceable market.
With the Reg A+ now live, investors can participate in Greenfield Robotics' growth story.
This Reg A+ offering is made available through StartEngine Primary, LLC, member FINRA/SIPC. Please read the Offering Circular and related disclosures before investing. This investment is speculative, illiquid, and involves a high degree of risk, including the possible loss of your entire investment.
Today:
OpenAI DevDay 2026 launches always-on dots, GPT-6.1 Sol, Space, Codex Cloud and more than 20 updates
Claude Sonnet 5.5 runs more than 30% faster and costs less for most work
Nvidia moves agent safety outside the model with OpenShell and Sentry
Grok Team Bots give teams shared AI coworkers with common context
Meta launches Enterprise Platform around Muse, business agents and developer tools
Dots were the headline launch. Powered by GPT-6 Astra, each dot is an always-on agent with its own cloud computer and browser, persistent context, connected tools and access to more than 4,000 apps through the plugin ecosystem. A dot can continue work in the background, follow up when it needs a decision, and stay available across ChatGPT, voice, Slack and Microsoft Teams, with text-message access planned later.
OpenAI is pairing dots with ChatGPT Space, a shared environment where teammates, ChatGPT and their agents work from the same files and project context. Space introduces Pages, editable documents for writing, research, charts, images and ongoing agent work; collaborative slides that export to PowerPoint and Google Slides are coming in the next few weeks, while Business and Enterprise customers are also getting Team Tasks, @ChatGPT inside Slack and Teams, and a Meetings plugin that saves summaries and action items into Space.
The new model launch is GPT-6.1 Sol, which OpenAI says approaches GPT-6 Astra on agentic coding, computer use and professional work while keeping standard pricing at $2 per million input tokens and $10 per million output tokens. Cached input drops to $0.10 per million tokens, the context window remains 1.05 million tokens, and the model is available in the API, ChatGPT Work and Codex. OpenAI also introduced Ultrafast processing, reaching up to roughly 300 tokens per second in Codex and up to eight times standard Astra speed, alongside a new $500-per-month Pro 500 plan with the company’s highest usage limits.
Codex received one of the largest groups of updates. Codex Cloud can now run coding tasks in reusable hosted environments from desktop, web or mobile and keep working while the user’s computer is asleep; the CLI adds voice input and an /agents view for tracking parallel tasks, the desktop app adds a dedicated code-review workflow, and Codex Security Cloud can scan entire GitHub repositories on demand or on a schedule and prepare fixes for review.
The developer platform is expanding in parallel. The Agents API gains computer use and multi-agent capabilities, a new Decisions API in limited preview uses a version of GPT-6 Luna for fast classification and routing among predefined choices, and OpenAI is working with Amazon on Bedrock Managed Agents so the same agent capabilities can run inside AWS. Plugin Extensions can now add their own sidebar surfaces and interactive panels inside ChatGPT, plugins can respond to external events through MCP Events, and Sign in with ChatGPT lets participating apps use a user’s ChatGPT allowance with permission.
OpenAI also announced enterprise privacy and purchasing tools. Private Intelligence is designed for sensitive workloads with zero data retention and automated safety review that OpenAI staff cannot inspect, while Private Inference based on confidential computing is planned for preview this fall. A new OpenAI Marketplace is launching in beta with 32 partners, letting eligible enterprise customers apply part of their OpenAI commitment toward qualifying software from companies including Adobe, Figma, Salesforce and CrowdStrike.
The rollout is split across products and plans. Dots are beginning with eligible Pro and Business Premium users, with Enterprise beta access controlled by administrators; Space is available on Pro, Business and Enterprise on web and desktop; GPT-6.1 Sol is already available in the API, Work and Codex; and several collaboration, privacy and speed features remain in preview or are scheduled for the coming weeks.
🧠RESEARCH
Active Taskless Distillation shows that post-training capabilities can transfer through task-unrelated outputs. A student learns only from single-word teacher choices, without target-task examples, logits, or parameters. In a Qwen2.5-1.5B coding experiment, the method improved HumanEval+ by 5.34 percentage points and also transferred science, commonsense, and reading skills across multiple models.
YuE2 unifies symbolic composition and full-song audio generation in one model. It first writes an editable melody-and-chord score, then converts that plan into vocals and accompaniment. On WildSongBench, best-of-eight generation reached a 6.96 SongBench average, while expert evaluations found symbolic planning improved overall musical quality and editability substantially in tests.
DN-MOPD addresses a weakness in multi-teacher distillation where one specialist’s feedback can dominate shared training. It rescales each domain’s token-level feedback by its measured spread, improving six-task averages across Qwen3.5 models at three sizes. Experiments show much of the gain comes from reducing overly strong instruction-following updates during training overall.
📲SOCIAL MEDIA
🗞️MORE NEWS
Claude Sonnet 5.5 is the second model in Anthropic’s 5.5 family and is positioned for well-scoped everyday work, coding fixes, documents, slides and spreadsheets. Anthropic says it runs more than 30% faster than Sonnet 5, costs up to 30% less for most tasks, and scores 70.6% on Terminal-Bench 4.0 while finishing only two points behind Opus 5.5 on GDPval-AA. API list pricing remains $2 per million input tokens and $10 per million output tokens, while Claude Haiku 5.5 is scheduled to join the family in the coming weeks.
Nvidia’s Open Agent Safety Platform combines the open-source OpenShell runtime with an independent hardware watchdog called Sentry to enforce agent boundaries outside the model itself. OpenShell isolates agents, verifies policies and controls files, networks, tools and credentials, while Sentry runs on BlueField-4 DPUs to monitor activity out of band and quarantine agents in milliseconds when they leave approved boundaries. The reference design is optimized for Nvidia Vera and BlueField systems but OpenShell is open source and can also be extended to third-party compute platforms.
Team Bots are shared Grok Bots built around a role or workflow so everyone on a team can work from the same files, instructions, apps and expertise. Each bot combines shared context, plugins, credentials and memory while keeping individual conversations and user-specific memories separate. Team Bots are available in public beta for Teams and Enterprise customers, with xAI describing internal uses including account briefings, engineering coordination and company-wide data questions.
Meta Enterprise Platform is a new business unit focused on bringing Meta’s AI models, agents and infrastructure to companies and developers. The initial stack includes Muse, Meta Business Agent, Muse API, Muse Code and related enterprise tools, with former MongoDB CEO Chirantan “CJ” Desai joining Meta as Chief Enterprise Platform Officer. The launch establishes a new enterprise pillar for Meta, although detailed product packaging, pricing and rollout schedules have not yet been announced.
Manus 2.0 introduces a new agent architecture called Cascade, which selectively loads specialized capabilities and in one tested setup used 23.2% fewer tokens, finished tasks 28.2% faster and cost 32% less than the previous system. The release adds dedicated Cloud Computers for always-on projects, event-triggered Automations, Manus Studio for professional creative work and software building, and Cue, a personal-agent app where agents can have their own identity, phone number, wallet and computer. Manus 2.0 is available now across web, desktop and mobile, while Cue is in early access with iOS pending App Store review.
Eleven v4 is ElevenLabs’ new expressive text-to-speech model, with v4 Turbo providing a lower-latency version for live voice agents across more than 90 languages. ElevenLabs says v4 ranks first on Artificial Analysis’ speech leaderboard and was preferred by roughly 75% of listeners in its own blind comparisons, while Turbo reaches about 100 milliseconds of median inference latency and roughly 150 milliseconds to first speech. Both models are available through ElevenCreative, ElevenAgents and the API, with temporary launch pricing of $0.022 per 1,000 characters for v4 and $0.011 for Turbo through October 12.
What'd you think of today's edition? |
Reply