Executive Summary
September 29 brought a dense mix of frontier-model releases, agent-platform upgrades, and U.S. policy changes. OpenAI released GPT-6.1 Sol with near-Astra positioning, lower pricing, a 1.05-million-token context window, and beta multi-agent support, while also adding computer use to the Agents API and an Ultrafast processing tier for GPT-6 Astra. In Washington, the White House ordered executive agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in future non-statutory communications, while the Associated Press reported that major technology companies signed a voluntary AI accord centered on internal and external reviews. Enterprise launches from Nasdaq, Honeycomb, and Braze showed the day’s broader commercial theme: governed AI agents are moving from pilots into financial operations, observability, and customer engagement.
Most Important News, Ranked by Significance
1. OpenAI releases GPT-6.1 Sol
OpenAI released GPT-6.1 Sol for complex coding, computer use, and professional work. The official model documentation describes near-Astra performance at a lower cost, with a 1,050,000-token context window and up to 128,000 output tokens. Standard API pricing is $2 per million input tokens, $0.10 for cached input, $2.50 for cache writes, and $10 for output for prompts up to 272,000 input tokens.
The release is notable not only for its price-performance claim but also for its risk profile. OpenAI’s September 29 system-card addendum classifies GPT-6.1 Sol as Critical for cybersecurity capability and High for biological and chemical capability under its Preparedness Framework. OpenAI says it applies the same safeguard stack used for GPT-6 Astra.
Why it matters: A model positioned close to the frontier at materially lower cost can accelerate adoption of advanced coding and agentic workflows, while the stated capability classifications underline the growing operational burden of deploying such systems safely.
2. The White House orders federal agencies to replace “AI” with “SI”
A September 29 executive order directs U.S. executive departments and agencies, to the maximum extent permitted by law, to use “Super Intelligence” and “SI” instead of “Artificial Intelligence” and “AI” in official correspondence, public communications, websites, reports, policy documents, and other non-statutory material. The order keeps the existing statutory definition of artificial intelligence for implementation purposes and asks the Assistant to the President for Science and Technology to submit proposed legislative language within 60 days.
Why it matters: The order changes federal terminology immediately even though it does not itself rewrite existing statutes, regulations, contracts, grants, or previous presidential actions. Agencies and contractors may need to account for a widening difference between statutory language and current executive-branch communications.
3. Major AI companies sign a voluntary review accord
The Associated Press reported on September 29 that President Donald Trump and leaders of major technology companies signed a voluntary accord intended to include internal and external reviews of AI systems. The available announcement framed the agreement as industry self-policing rather than a mandatory regulatory regime; further implementation details were still expected.
Why it matters: The accord places participating companies at the center of oversight during a period of heightened concern about advanced-model behavior and security. Its practical significance will depend on the review standards, independence of outside assessments, disclosure rules, and consequences for noncompliance.
4. Nasdaq launches an agentic AI framework for capital markets
Nasdaq Calypso introduced a governed operating environment for financial institutions to build, connect, and scale AI agents across trade and treasury workflows. The framework includes a natural-language assistant for analysis of system-of-record data and is designed to preserve Calypso’s existing governance and controls as institutions expand their agentic workforce.
Why it matters: Capital-markets infrastructure has strict auditability and control requirements. Embedding agents inside a governed system of record is a more consequential step than adding a general-purpose chatbot because it moves agentic automation closer to regulated operational workflows.
5. Cloudflare publishes frontier-model testing against its web application firewall
Cloudflare reported results from using frontier AI models to test its own web application firewall. The company’s central observation is that advanced models can rapidly iterate and mutate attack payloads in response to live feedback, including changing encodings, request locations, and vulnerability targets.
Why it matters: The work highlights a concrete defensive challenge: AI-assisted attackers can explore variations faster than traditional manual testing. Security teams may need evaluation methods that measure adaptive, multi-step attack behavior rather than relying only on static payload sets.
Notable New Tools and Updates
OpenAI Agents API gains computer use
Developers can now let Agents API workloads operate software through an OpenAI-hosted browser. The official changelog says website-access approvals and sign-in are handled by the developer’s application.
Beta multi-agent support arrives with GPT-6.1 Sol
The Responses API can let GPT-6.1 Sol create and coordinate subagents for parallel work, then synthesize their results. OpenAI’s documentation identifies codebase exploration, documentation, and implementation as representative workloads.
GPT-6 Astra gets an Ultrafast tier
OpenAI added an Ultrafast service tier for GPT-6 Astra in the Responses API. It is designed to reduce time between generated tokens and is offered with global processing or U.S. data residency, subject to availability and rate limits.
Codex Cloud expands for enterprise teams
OpenAI’s September 29 Enterprise and Edu release notes say authorized members can share reusable cloud environments and continue Codex tasks from desktop, web, or mobile in isolated workspaces, including while a local computer is asleep. Workspace administrators can also configure approved app connections for team use.
OpenAI Marketplace enters beta
OpenAI announced a beta program allowing eligible enterprise customers to apply part of their OpenAI commitment toward qualifying partner software. Eligibility is determined by customer, product, and purchase.
Honeycomb launches an AI observability suite
Honeycomb introduced AI Ecosystem, with Agent Fleet Performance, LLM Cost Tracking, and Agent Conversations, plus new Canvas MCP connectors. The launch targets fleet-wide visibility into agent health, cost, and individual conversations.
Braze adds decisioning and agentic marketing tools
Braze unveiled Decisioning Studio Go, Agentic Standards for campaign quality assurance, Operator Connect for headless workflows, and expanded conversational AI capabilities across messaging channels.
Original Sources
Coverage date: September 29, 2026. Publication dates were checked against the linked source pages. Duplicate developments were consolidated.
September 29 brought a dense mix of frontier-model releases, agent-platform upgrades, and U.S. policy changes. OpenAI released GPT-6.1 Sol with near-Astra positioning, lower pricing, a 1.05-million-token context window, and beta multi-agent support, while also adding computer use to the Agents API and an Ultrafast processing tier for GPT-6 Astra. In Washington, the White House ordered executive agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in future non-statutory communications, while the Associated Press reported that major technology companies signed a voluntary AI accord centered on internal and external reviews. Enterprise launches from Nasdaq, Honeycomb, and Braze showed the day’s broader commercial theme: governed AI agents are moving from pilots into financial operations, observability, and customer engagement.
Most Important News, Ranked by Significance
1. OpenAI releases GPT-6.1 Sol
OpenAI released GPT-6.1 Sol for complex coding, computer use, and professional work. The official model documentation describes near-Astra performance at a lower cost, with a 1,050,000-token context window and up to 128,000 output tokens. Standard API pricing is $2 per million input tokens, $0.10 for cached input, $2.50 for cache writes, and $10 for output for prompts up to 272,000 input tokens.
The release is notable not only for its price-performance claim but also for its risk profile. OpenAI’s September 29 system-card addendum classifies GPT-6.1 Sol as Critical for cybersecurity capability and High for biological and chemical capability under its Preparedness Framework. OpenAI says it applies the same safeguard stack used for GPT-6 Astra.
Why it matters: A model positioned close to the frontier at materially lower cost can accelerate adoption of advanced coding and agentic workflows, while the stated capability classifications underline the growing operational burden of deploying such systems safely.
2. The White House orders federal agencies to replace “AI” with “SI”
A September 29 executive order directs U.S. executive departments and agencies, to the maximum extent permitted by law, to use “Super Intelligence” and “SI” instead of “Artificial Intelligence” and “AI” in official correspondence, public communications, websites, reports, policy documents, and other non-statutory material. The order keeps the existing statutory definition of artificial intelligence for implementation purposes and asks the Assistant to the President for Science and Technology to submit proposed legislative language within 60 days.
Why it matters: The order changes federal terminology immediately even though it does not itself rewrite existing statutes, regulations, contracts, grants, or previous presidential actions. Agencies and contractors may need to account for a widening difference between statutory language and current executive-branch communications.
3. Major AI companies sign a voluntary review accord
The Associated Press reported on September 29 that President Donald Trump and leaders of major technology companies signed a voluntary accord intended to include internal and external reviews of AI systems. The available announcement framed the agreement as industry self-policing rather than a mandatory regulatory regime; further implementation details were still expected.
Why it matters: The accord places participating companies at the center of oversight during a period of heightened concern about advanced-model behavior and security. Its practical significance will depend on the review standards, independence of outside assessments, disclosure rules, and consequences for noncompliance.
4. Nasdaq launches an agentic AI framework for capital markets
Nasdaq Calypso introduced a governed operating environment for financial institutions to build, connect, and scale AI agents across trade and treasury workflows. The framework includes a natural-language assistant for analysis of system-of-record data and is designed to preserve Calypso’s existing governance and controls as institutions expand their agentic workforce.
Why it matters: Capital-markets infrastructure has strict auditability and control requirements. Embedding agents inside a governed system of record is a more consequential step than adding a general-purpose chatbot because it moves agentic automation closer to regulated operational workflows.
5. Cloudflare publishes frontier-model testing against its web application firewall
Cloudflare reported results from using frontier AI models to test its own web application firewall. The company’s central observation is that advanced models can rapidly iterate and mutate attack payloads in response to live feedback, including changing encodings, request locations, and vulnerability targets.
Why it matters: The work highlights a concrete defensive challenge: AI-assisted attackers can explore variations faster than traditional manual testing. Security teams may need evaluation methods that measure adaptive, multi-step attack behavior rather than relying only on static payload sets.
Notable New Tools and Updates
OpenAI Agents API gains computer use
Developers can now let Agents API workloads operate software through an OpenAI-hosted browser. The official changelog says website-access approvals and sign-in are handled by the developer’s application.
Beta multi-agent support arrives with GPT-6.1 Sol
The Responses API can let GPT-6.1 Sol create and coordinate subagents for parallel work, then synthesize their results. OpenAI’s documentation identifies codebase exploration, documentation, and implementation as representative workloads.
GPT-6 Astra gets an Ultrafast tier
OpenAI added an Ultrafast service tier for GPT-6 Astra in the Responses API. It is designed to reduce time between generated tokens and is offered with global processing or U.S. data residency, subject to availability and rate limits.
Codex Cloud expands for enterprise teams
OpenAI’s September 29 Enterprise and Edu release notes say authorized members can share reusable cloud environments and continue Codex tasks from desktop, web, or mobile in isolated workspaces, including while a local computer is asleep. Workspace administrators can also configure approved app connections for team use.
OpenAI Marketplace enters beta
OpenAI announced a beta program allowing eligible enterprise customers to apply part of their OpenAI commitment toward qualifying partner software. Eligibility is determined by customer, product, and purchase.
Honeycomb launches an AI observability suite
Honeycomb introduced AI Ecosystem, with Agent Fleet Performance, LLM Cost Tracking, and Agent Conversations, plus new Canvas MCP connectors. The launch targets fleet-wide visibility into agent health, cost, and individual conversations.
Braze adds decisioning and agentic marketing tools
Braze unveiled Decisioning Studio Go, Agentic Standards for campaign quality assurance, Operator Connect for headless workflows, and expanded conversational AI capabilities across messaging channels.
Original Sources
- OpenAI API changelog — September 29, 2026
- GPT-6.1 Sol model documentation
- GPT-6.1 Sol system-card addendum — September 29, 2026
- OpenAI multi-agent guide
- ChatGPT Enterprise and Edu release notes — September 29, 2026
- White House executive order — September 29, 2026
- Associated Press report on the voluntary AI accord — September 29, 2026
- Nasdaq Calypso agentic AI framework announcement — September 29, 2026
- Cloudflare frontier-model WAF testing — September 29, 2026
- Honeycomb AI observability announcement — September 29, 2026
- Braze AI portfolio announcement — September 29, 2026
Coverage date: September 29, 2026. Publication dates were checked against the linked source pages. Duplicate developments were consolidated.