OpenAI Release Notes

Promoting transparency within the limits of today’s technology.

Content provenance helps people understand where content came from, how it was created or edited, and whether it contains signals associated with our models. We’ve already made tools publicly available to identify images and audio generated by our models. Today, we’re sharing our approach to text watermarking in response to the EU AI Act, and how it fits into our broader work.

The EU AI Act requires generative AI providers to make generated text identifiable in a machine-readable way. Text watermarking and detection remain early technologies with significant limitations, and views about their benefits and responsible uses are still developing. Our phased approach reflects both the EU AI Act requirements as well as the technology’s limitations, with an emphasis on transparency about what a text watermark can and cannot tell people:

• Starting today, API customers globally will be able to opt in to text watermarking for select models. Text watermarking will remain off by default in the API.
• Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union.
• We’re opening applications to access our text watermark detector. Access will initially be limited to approved researchers and expert organizations that can help us evaluate and improve the technology.
  • The above only applies to text provenance—our verification tools for audio and images, including our openai.com/verify⁠ web tool and our Content Provenance API⁠ (opens in a new window), will continue to be publicly accessible to organizations looking to understand whether an image or audio file was generated by one of our systems.

How our watermarking works and performs

Our text watermarking technology, textGrain, adds an invisible statistical signal to the model’s word choices. Our detector looks for that signal to assess whether a passage contains an OpenAI watermark. More details about how textGrain works can be found in our technical report⁠ (opens in a new window), which will be updated with additional details in the coming weeks. We also plan to make the technology available in open source so that others can build on it.

In our evaluations, textGrain matched or exceeded the performance of other approaches we tested, including SynthID for text. Even so, strong performance under ideal conditions does not guarantee reliable detection in everyday use.

Detectors can make two kinds of errors: they can report a watermark where none is present—a false positive—or miss a watermark that is present—a false negative. Our evaluations below illustrate some of the challenges:

• Shorter or more constrained text is harder to detect.
  At a target false positive rate of 1%, our detector identified watermarks in about 80% of 200-token passages, compared with about 95% of 400-token passages, for content such as psychology. Detection rates were substantially lower for content such as mathematics, where there is less flexibility in word choice.
• Editing can weaken the watermark.
  In an evaluation of 400-token passages, replacing 10% of words with synonyms reduced detection from about 92% to 66%. Replacing 25% of words reduced it to 17%.

These limitations contribute to our decision to provide initial detector access only to approved researchers and expert organizations, who can help us evaluate reliability and responsible uses.

Impact of watermarking on output quality

Across the benchmarks we use to assess Astra, our latest frontier model, we do not see meaningful performance differences with and without watermarking.

What a text watermark doesn’t tell you

Text watermarks provide a limited signal about the role our systems played in a passage. It’s important to understand what can and cannot be concluded from a detection result.

• A watermark does not measure human contribution.
  It can indicate that an OpenAI system generated or processed part of a passage, but not how much human judgment, editing, or creativity went into it.
• A watermark does not establish ownership or responsibility.
  It does not determine who owns the text, whether its use was lawful, whether disclosure was required, or who is responsible for it.
• A watermark does not identify the user.
  It does not associate a person, organization, account, prompt, or conversation with the text.
• A watermark does not verify accuracy.
  It does not tell you whether a passage is true, misleading, harmful, or presented in the right context.
• The absence of a detected watermark does not prove human authorship.
  Text generated with OpenAI tools may be too short, edited, or translated for detection to work reliably. It may also come from an unsupported model, predate watermarking, or have been generated by another company’s tools.

Our next steps

• Rolling out text watermarking in the EU.
  Over the coming weeks, we will introduce text watermarking to eligible ChatGPT and Codex users across all plans in the EU only. We are not making text watermarking a global default at launch. This regional approach gives us room to learn from real-world use and feedback.
• Enabling opt-in watermarking for API customers.
  Starting today, API customers around the world will be able to opt in to watermarked text outputs for select models. This lets customers decide how watermarking fits their transparency obligations and the experiences they provide to users. We are also working with cloud partners to make watermarking available for OpenAI model outputs accessed through their services in the coming weeks.
• Providing detector access to researchers and expert organizations.
  Approved researchers and expert organizations can apply starting today. In accordance with the Code of Practice⁠ (opens in a new window), access will be initially granted on a case-by-case basis to support evaluation and improvement of text provenance. The tool will report whether it detects an OpenAI watermark, without identifying the user or revealing their prompts or conversations. Given the risk of missed watermarks and false positives, we are not making it publicly available at launch.

We expect to revisit each part of this approach as the technology, standards, and evidence evolve.

Our broader approach to content provenance

No single provenance technique is enough on its own, so we take a layered approach that combines open standards, durable watermarking, and verification tools.

We add Content Credentials to supported image outputs, are C2PA conformant⁠ (opens in a new window), embed invisible SynthID⁠ (opens in a new window) watermarks in supported images and audio, and make image and audio verification available through openai.com/verify⁠ and our Content Provenance API⁠ (opens in a new window). These techniques complement one another: Content Credentials can record a file’s origin and history, while invisible watermarks can preserve a signal when metadata is removed.

As we described in our election safeguards work, provenance can help people and platforms assess potentially misleading AI-generated content, including deepfakes. We pair these signals with policies, abuse detection, reporting, investigations, and enforcement, and work with researchers, standards bodies, platforms, and civil society to make provenance useful across the wider ecosystem.

Extending provenance to text requires accounting for how easily it can be rewritten, translated, or edited. As we discussed in June, our approach must reflect the practical limits of current technology. We’ll continue improving detection, studying how watermarks withstand editing and translation, and exploring ways to distinguish AI assistance from AI authorship more meaningfully. We’ll adapt our approach as evidence, standards, and regulatory requirements evolve, and expand detector access when we believe results can be interpreted responsibly.

For more information, please visit our help center article⁠ (opens in a new window).

It looks like the OpenAI team decided that the flood of DevDay announcements was not enough. 😄

Starting today, there will be one clear improvement each day for most of all Codex or Work users. On days without one, everyone gets a full reset instead.

Because why not?

We’ll keep this topic updated each day.

Day 1

GPT-6 Astra and GPT-6.1 Sol should now feel around 50% faster by default when used through your subscription.

This applies across OpenAI products and partners that support Sign in with ChatGPT, including OpenCode, Pi, Amp, and Devin.

There is nothing you need to change. You should notice the improvement within the next two hours.

Source

OpenAI has shared how it plans to implement text provenance under the EU AI Act.

Starting today, API customers worldwide can opt in to text watermarking for select models. Over the coming weeks, OpenAI will also add an invisible watermark to eligible ChatGPT and Codex text generated in the EU. It will remain off by default in the API and will not become a global default at launch.

The technology, called textGrain, adds a statistical signal through the model’s word choices. A detector can look for this signal, but short passages and later edits make it less reliable. This is why detector access will initially be limited to approved researchers and expert organizations.

The watermark can indicate that an OpenAI system generated or processed part of a text. It cannot tell who used the system, how much a person contributed, who owns the text, or whether the content is accurate. If no watermark is detected, that also does not prove that a person wrote it.

Read the announcement in full

Bringing products and services to life

A new visual ad format helps people explore products and services, with expanded measurement to help businesses understand ad impact.

We’re introducing a new visual ad format in ChatGPT, expanding our measurement tools and partnerships, and developing new ways to understand brand suitability.

ChatGPT reaches 1.2 billion people each week, making it the world’s largest AI-native consumer platform. Advertising helps us bring the benefits of AI to more people. That broad access is central to our mission, and a strength of our business. It allows us to serve a much larger market than subscriptions alone, while giving businesses a new way to reach people as they explore ideas, make decisions and discover products and services relevant to what they’re trying to accomplish.

Bringing products and services to life

Our new visual ad format helps people imagine how products and services could fit into their lives. Through images showing product inspiration, product usage, or the experiences they make possible, people can discover a new brand, understand what makes an offering distinctive, and decide what to buy. For advertisers, this creates new opportunities to tell their story through images.

Initially, we’ll test this new ad format during image generation in ChatGPT. Ads will be clearly labeled, and remain separate from the image being created. As with our other ad formats, advertising does not influence the answers ChatGPT provides.

Testing will begin later this month in the US with an initial group of advertisers.

Measuring the impact of ChatGPT Ads

Our measurement philosophy for ChatGPT Ads is focused on supporting both in-house and partner-led measurement solutions. Today, we’re expanding our measurement tools and partner ecosystem to help advertisers understand how ChatGPT Ads contribute to their business using the systems they already rely on.

Integrations with Hightouch, Tealium, and LiveRamp make it easier for advertisers to send conversion data from their existing systems to ChatGPT Ads.

We now support leading attribution partners across the web and app space, enabling Conversions API, reporting, and click attribution solutions with companies including AppsFlyer, Triple Whale, Adjust, DV Rockerbox, Northbeam, Branch, Singular, Kochava, Airbridge, and Tenjin. We also have partnerships with full-funnel and advanced measurement partners, including Fospha, Measured, and INCRMNTAL.

Incrementality measurement plays an important role in building a fuller understanding of advertising’s business impact. While our work on incrementality is still in its early stages, we’re partnering with Haus, Measured, and WorkMagic to explore geo-based experiments to help advertisers understand the causal impact of ChatGPT Ads.

Early partner findings illustrate strong performance across different measurement approaches. According to DV Rockerbox, WeightWatchers’ attributed cost per acquisition on ChatGPT Ads was 15.3% lower than its blended paid-search benchmark. WorkMagic reported statistically significant lift for wellness brand Dose, with 67% of incremental purchases coming from net-new customers. And according to Triple Whale, 93% of Portland Leather’s visitors from ChatGPT Ads were new.

Read more about these measurement tools and partnerships on our Ads Blog here (opens in a new window).

Building confidence in alignment with our Ads Principles

Confidence in advertising also depends on understanding the environment where ads can appear. OpenAI’s advertising policies and guardrails are designed to protect users, preserve trust, and give brands confidence in their ads placement. Our placement guardrails assess whether a conversation is appropriate for advertising, helping keep ads out of emotionally vulnerable, sensitive, or otherwise unsuitable contexts.

Automated review, human oversight, and ongoing monitoring help us identify and address policy violations and improve our safeguards over time. For narrow placement requirements specific to a brand’s own policies, Negative Phrases are now available to qualifying advertisers.

We’re also developing brand suitability evaluation pilots with DoubleVerify (DV) and Integral Ad Science (IAS) to help advertisers build a better understanding of the ads environment. This approach allows independent partners to assess how our brand safety standards are applied in a controlled testing environment, without accessing private user conversations. Our goal is to give advertisers greater transparency into how those safeguards work while protecting conversation privacy.

“As advertising expands into AI-native environments, marketers need confidence that brand safety protections are durable, consistent, and built to operate at scale. IAS’s independent evaluation will assess how OpenAI’s safeguards perform across a broad and evolving range of scenarios, giving advertisers greater transparency into brand safety and suitability, while preserving the privacy of real user conversations.”
— Lidiane Jones, CEO of Integral Ad Science

“ChatGPT is creating a fundamentally different advertising environment, where context is defined by the conversation itself rather than a page or video. As advertising in these environments scales, delivering ads next to conversations that align with advertisers’ suitability goals becomes essential. We’re excited to partner with OpenAI with the goal to bring independent measurement to this dynamic environment while respecting user privacy, helping to give advertisers greater confidence as they explore this emerging and promising channel.”
— Mark Zagorski, CEO of DoubleVerify

Expanding the features of ChatGPT Ads

Our new visual ad format helps people discover products and services and imagine how they could fit into their lives, while giving advertisers new ways to tell their stories more visually. Across ChatGPT Ads, expanded measurement and brand suitability efforts help businesses understand campaign impact and build confidence in the contexts where their ads appear. These features are consistent with our ads principles: ChatGPT’s answers stay independent, conversations stay private, and users remain in control of their experience.

Today’s updates are part of our broader ambition to help people discover products and services, evaluate options, and make decisions in ChatGPT. We will continue to evolve our ad platform with new formats and capabilities to help businesses reach customers in useful and relevant ways and provide opportunities for businesses of all sizes to grow and thrive.

Businesses can sign up today for ads at ads.openai.com (opens in a new window).

New Features

• Browse older tasks in the agent command center with a keyboard-accessible “Show more” action. (#49106)
• Select transcript text and paste with middle-click in fullscreen mode on supported local Linux X11 terminals. (#49112)
• Start sessions outside a project with workspace defaults when policy permits, and restore saved permissions when resuming. (#49160)
• Added opt-in Guardian review capabilities to retrieve earlier user instructions and include context from agent handoffs. (#49036, #49057)

Bug Fixes

• Unsent queued messages now resume after reconnection once uncertain submissions are resolved, avoiding duplicate sends. (#49105)
• The terminal UI now preserves server provider, reasoning-summary, and verbosity settings and shows the correct sessions in resume and fork history. (#49144, #49161, #49171)
• Fixed Windows sandbox PowerShell fallbacks and long-path permission repairs, and suppressed unwanted console windows from background helpers. (#49019, #49058, #49098, #49164, #49386)
• Subagents now retain environments that are still starting and receive their configuration or preparation failure. (#49075)
• Prevented SQLite stalls during connection setup and logging, and surfaced initialization errors instead of masking them as timeouts. (#49032, #49102)
• Explicit provider model catalogs no longer include unsupported bundled models or reuse stale entries after refresh failures. (#49135)

Documentation

• Clarified how provider credentials use the configured storage backend and how env_key identifies the API-key environment variable. (#49118)

Chores

• Reduced repeated plugin-loading work by caching parsed manifests and reusing HTTP connections for remote plugin requests. (#49099, #49100)
• Added background reclamation of unused log database space to reduce disk usage. (#49069)

Changelog

Full Changelog: rust-v0.159.0...rust-v0.160.0

• #48983 Avoid full metadata rewrites for thread timestamp updates @jif-oai
• #49000 Isolate the memory startup metadata test from Git enrichment @jif-oai
• #49019 Use a compatible PowerShell fallback for the Windows MXC sandbox @iceweasel-oai
• #49028 Use the numeric ioctl value in the macOS sandbox policy @aionescu-oai
• #49031 Clarify ChatGPT sign-in success copy @bc-openai
• #49032 Avoid SQLite stalls from connection setup and stderr span logging @jif-oai
• #49036 Add opt-in conversation history retrieval to Guardian reviews @felixxia-oai
• #49037 Show the Plan mode cycling hint in the fullscreen status line @fcoury-oai
• #49038 Preserve encrypted agent messages in Guardian reviews @felixxia-oai
• #49041 Copy selections within inline code as plain text @fcoury-oai
• #49043 Update Pro plan display names in the TUI @etraut-openai
• #49057 Add handoff-aware root context for Guardian reviews @jif-oai
• #49058 Fix Windows sandbox ACL repair for long runtime paths @darius-oai
• #49060 Update Guardian messaging snapshots for native agent messages @dkovalenko-oai
• #49065 Update Guardian handoff snapshot for separate agent messages @cassirer-openai
• #49067 Keep configuration values out of Windows sandbox policy events @zm-oai
• #49069 Reclaim unused SQLite log database pages in the background @dkovalenko-oai
• #49073 Surface realtime voice catalog failures in the TUI @etraut-openai
• #49074 Propagate Cargo package versions to Bazel Rust targets @tamird
• #49075 Preserve pending environments when spawning subagents @sayan-oai
• #49076 Avoid collecting unused Git metadata in skill analytics @tamird
• #49079 Update and centralize TUI subscription labels @etraut-openai
• #49082 Skip remote Git discovery for Guardian diff paths @jif-oai
• #49084 Track app-server running turns incrementally @tamird
• #49089 Render follow-up directive labels in the TUI and copied responses @etraut-openai
• #49093 Simplify startup promotions to platform-specific desktop app tips @etraut-openai
• #49096 Update h2 from 0.4.16 to 0.4.19 in Cargo and Bazel lockfiles @cassirer-openai
• #49097 Notify lifecycle extensions of compaction usage limits @xli-oai
• #49098 Resolve Windows sandbox PowerShell fallbacks on the exec server @zm-oai
• #49099 Cache parsed plugin manifests across plugin workflows @dkovalenko-oai
• #49100 Reuse the HTTP connection pool for remote plugin requests @dkovalenko-oai
• #49102 Preserve SQLite vacuum modes and surface pool initialization errors @bc-openai
• #49103 Balance Windows Bazel test shards using duration estimates @jgershen-oai
• #49105 Resume unsent TUI input after reconnecting @etraut-openai
• #49106 Add history pagination to the agent command center @bc-openai
• #49112 Add X11 primary selection and middle-click paste support @fcoury-oai
• #49114 Point remote compaction tests at the mock ChatGPT server @OLI-OAI
• #49117 Attribute analytics requests to each thread's product SKU @papayo-oai
• #49118 Correct provider authentication storage documentation @celia-oai
• #49119 Add recovery guidance to content-filter retries @won-openai
• #49127 Deduplicate cloud and executor skill listings before budgeting @TAFOYA-OAI
• #49130 Move content-filter guidance into the shared Responses retry handler @won-openai
• #49135 Treat explicit provider model catalogs as authoritative @andrew-gu-oai
• #49136 Remove the plus separator after Option symbols in TUI key hints @bc-openai
• #49138 Expose original error details to turn lifecycle contributors @bryanashley
• #49144 Preserve server reasoning summary and verbosity settings in the TUI @etraut-openai
• #49145 Hide reasoning summary settings in /status for server connections @etraut-openai
• #49147 Simplify cloud task base URL normalization @akira-oai
• #49153 Omit blockquote markers when copying quoted selections in the TUI @bc-openai
• #49160 Support projectless TUI sessions with workspace defaults @etraut-openai
• #49161 Honor app-server provider defaults in the TUI @etraut-openai
• #49164 Suppress Windows console windows for background subprocesses @etraut-openai
• #49171 Fix model provider lookup for TUI history @etraut-openai
• #49386 [0.160] Backport remaining Windows console fix to frozen alpha.6 @andrewguoai
• #49763 [0.160] Backport maintenance-line catalog and security reminder updates @andrewguoai

Manage plugins and marketplaces in Admin console

Workspace owners and admins in ChatGPT Enterprise plans can now manage plugins and marketplaces in Admin console. Select your ChatGPT workspace, then open Plugins to manage plugins or Marketplaces to manage your workspace’s plugin marketplaces.

Apps remain available in Workspace settings > Apps. You can also open the Apps page by selecting Manage Legacy Apps from the Plugins page in Admin console. Existing workspace permissions and app access controls continue to apply. Learn more about managing plugins and apps.

Manage plugins and marketplaces in Admin console

Workspace owners and admins in ChatGPT Business can now manage plugins and marketplaces in Admin console. Select your ChatGPT workspace, then open Plugins to manage plugins or Marketplaces to manage your workspace’s plugin marketplaces.

Apps remain available in Workspace settings > Apps. You can also open the Apps page by selecting Manage Legacy Apps from the Plugins page in Admin console. Existing workspace permissions and app access controls continue to apply.

Learn more about managing plugins and apps.

See how clothes and accessories could look on you and save your favorites for later.

• Virtually try on clothing and accessories: A “Try on” button will appear on product listings for clothes and accessories in ChatGPT. Select it, then take or upload a selfie to generate a virtual try-on with ChatGPT Images. Your reference photo is saved for future try-ons. You can change or delete it in Settings → Personalization → Reference photos.
• Save your finds: Save any product to Favorites or create a folder to organize your finds in your ChatGPT Library.

Available in ChatGPT on mobile and web.

It’s now easier to bring your notes and documents into ChatGPT on your phone. Capture multiple pages in a row, and ChatGPT automatically combines them into a single PDF, ready to upload to your chat.

To get started, open the camera in ChatGPT, tap the three-dot menu, and select Scan. Rolling out on iOS.

Albertsons Cos. is expanding its work with OpenAI to improve the retail experience for millions of customers.

Retail runs on small decisions made at enormous scale: which products to promote, how to help an associate, what a developer should build next—and, closer to home, what’s for dinner. Grocery shopping is one of the most ordinary ways we take care of ourselves and the people around us. At Albertsons Cos., these decisions play out across more than 2,200 stores, including Albertsons®, Safeway®, Vons®, Jewel-Osco®, Shaw’s®, ACME® and Tom Thumb®, serving over 36 million customers each week.

OpenAI is working with Albertsons Cos. as the retailer continues to bring AI into everyday workflows, including ChatGPT Enterprise, as well as custom AI-powered experiences for its customers, data-driven recommendations and promotional insights built with the OpenAI API. The new Safeway experience in ChatGPT⁠ (opens in a new window) brings that innovation to millions of shoppers, helping them move from a dinner idea to checkout all in ChatGPT.

Saving teams time and accelerating retail innovation

Digital shopping, store operations, fulfillment, merchandising and customer experience all depend on systems that teams continually build, test and improve. Albertsons Cos. is expanding its use of AI across the organization to help teams navigate complexity, accelerate workflows and move promising ideas toward production faster. The work connects the technology behind a shopping trip to practical decisions: which products to promote, how to explain a recommendation and how to help customers find ingredients for dinner.

That rollout is part of a broader investment in how work gets done. Through ChatGPT Enterprise and other AI-powered tools, select teams are exploring new ways to streamline work, improve decision-making and identify practices that can scale across the organization.

ChatGPT Enterprise and OpenAI-powered applications are being used by teams building retail technology and by business functions where AI can drive meaningful impact. By developing capabilities in focused areas first, Albertsons Cos. can evaluate what works, build repeatable practices and create a strong foundation for broader innovation over time.

Bringing Safeway grocery shopping into ChatGPT

Shoppers are more open to using AI for groceries than for other types of purchases, according to dunnhumby’s February 2026 United States Consumer Trends Tracker⁠ (opens in a new window).

The new Safeway experience in ChatGPT is an example of how AI can make grocery shopping easier. Shoppers can start with a recipe, photo, digital list or simple request: “Plan pizza night for four,” “Put together a quick dinner from the deli,” or “Restock breakfast foods and snacks using store brands.” ChatGPT surfaces relevant products and savings, helps build a cart and directs customers to Safeway to check out.

“This is another practical way we are using AI to reduce friction and make everyday shopping easier,” said Jill Pavlovich, SVP, Digital Customer Experience at Albertsons Companies.

The new Safeway experience in ChatGPT expands on Albertsons’ AI shopping assistant and Ask AI capabilities and is planned to grow beyond Safeway to other grocery brands, including Albertsons, Vons, Jewel-Osco, Shaw’s, ACME and Tom Thumb.

A 360-degree partnership built to scale across retail

Albertsons Cos. is expanding its partnership with OpenAI to help teams work more efficiently and make grocery shopping easier for customers. Through the OpenAI API, Albertsons Cos. continues to explore how AI can create value on both sides of retail. For customers, that means more personalized assistance with building baskets, repurchasing familiar items and planning recipes and meals tailored to preferences and information provided by the customer.

Behind the scenes, Albertsons Cos. is combining predictive models with generative AI to deliver explainable, data-driven recommendations and promotional insights, helping merchants evaluate business considerations and make more informed decisions. That progress is supported by a broader set of OpenAI-powered capabilities, from employee tools to commerce apps and advertising solutions that can be extended across the enterprise.

Across more than 2,200 stores serving over 36 million customers each week, those improvements connect: teams build faster, decisions get better and putting food on the table becomes a little easier.

GPT-6.1 Sol is one of the very interesting Dev Day releases for the API because of the price performance developers get from this model.

GPT-6.1 Sol is getting close to Astra on complex work while costing less. The DevDay roundup puts that into perspective: improved coding and computer-use performance, with standard token prices at one-fifth of Astra’s.

Compared with the original GPT-6 Sol, GPT-6.1 Sol keeps the same $2/M input and $10/M output pricing, while cutting cached input from $0.20/M to $0.10/M.

On DeepSWE v1.1, GPT-6.1 Sol achieves 75.2% at high reasoning effort, surpassing GPT-6 Sol’s best score of 68.8% at maximum effort, at approximately 76% lower cost per task.

On AutomationBench, GPT-6.1 Sol scores 31.7% at medium reasoning effort, up 4.8 percentage points from GPT-6 Sol at the same setting.

On OSWorld 2.0’s offline set, it scores 71.4% versus Astra’s 73.5%, both at maximum reasoning effort, at roughly one-seventh of Astra’s cost per task.

For your most demanding work, choose Astra when maximizing quality matters most.

For complex work you want to run more often, GPT-6.1 Sol gives you near-Astra performance at a substantially lower cost.

GPT-6.1 Sol is available today in ChatGPT Work, Codex, and in the API.

Our premium speed tier, Ultrafast offers up to 8x faster token generation (300 tokens per second) in Codex and up to 6x in the API.

In Codex, Ultrafast generates code faster, so you can move more quickly from idea to code to iteration :racing_car:

In the API, combine it with WebSocket mode in the Responses API to build real-time experiences with frontier intelligence.

Ultrafast is available for Astra in Codex and ChatGPT Work on Enterprise plans, and through the new Pro 500 plan.

Ultrafast for Astra is also available to all developers via the API.

Ultrafast for GPT-6.1 Sol is coming soon.

Get access to Ultrafast in Codex and ChatGPT Work with the new Pro 500 plan.

Use Ultrafast mode with GPT-6 Astra over WebSockets, with SDK examples and an HTTP alternative.