ai lab
Also indexed as open-ai · chatgpt
OpenAI
OpenAI's importance comes from the combination of frontier research, a large developer ecosystem, and direct consumer distribution. A change to its models or products can alter how millions of people understand what AI software should do.6,3
50 tools · 45 Intel
Profile
Overview
From research lab to platform
OpenAI was founded in 2015 as an artificial intelligence research organization. It developed the GPT family of language models, then became a major consumer and developer platform through ChatGPT, its API, and product lines for reasoning, coding, image, audio, and video generation.1,2
ChatGPT changes the interface
The company helped move large language models from research demonstrations into everyday software. ChatGPT established conversational access as the dominant interface for general-purpose models, while the API made the same systems available to startups and established software companies.3
Research and product under one roof
OpenAI now operates as both a frontier research lab and a product company. That combination gives it unusual influence over model capabilities, developer conventions, and consumer expectations, but also makes leadership continuity, safety policy, and infrastructure execution central to any assessment of the lab.6,4
Governance is part of the story
Its structure has changed alongside that expansion. The nonprofit OpenAI Foundation governs the for-profit OpenAI Group public benefit corporation, an arrangement intended to combine mission control with access to the capital required for frontier model development. That structure, and the departure of several senior operators and researchers, makes governance part of the product story rather than a separate corporate footnote.6,8,9
Company evidence
Epoch AI dataset ↗Reported estimates, not audited figures. Confidence labels and source links are preserved from the dataset.
Reporting and context
- OpenAI Model SpecA primary document for understanding the behavior, priorities, and tradeoffs OpenAI intends its models to follow.
- OpenAI Codex lead on the new shape of product workFirst-hand context on how agentic coding is changing product development inside OpenAI.
- The future of software development with OpenAI's Sherwin WuOperational evidence about parallel agents, faster review, and the lab's own engineering practice.
- From GPT-2 to gpt-ossA technical account of how OpenAI's model architecture and open-weight strategy evolved.
Notable contributions
- 01Generative pre-training at product scaleThe GPT line helped establish large pretrained transformers as a general-purpose interface for language work.2
- 02Conversational AI as a mass-market productChatGPT turned model interaction into a mainstream consumer behavior and accelerated the current assistant and agent market.3
- 03Reasoning and coding model product linesThe o-series and Codex lines made test-time reasoning and agentic coding distinct product categories rather than hidden implementation details.4,5
Sources · 9+−
- 1Introducing OpenAIOpenAI · primary · Dec 11, 2015 ↗
- 2Improving Language Understanding by Generative Pre-TrainingOpenAI · paper · Jun 11, 2018 ↗
- 3Introducing ChatGPTOpenAI · primary · Nov 30, 2022 ↗
- 4Learning to reason with LLMsOpenAI · primary · Sep 12, 2024 ↗
- 5Introducing CodexOpenAI · primary · May 16, 2025 ↗
- 6About OpenAIOpenAI · primary ↗
- 7OpenAI executive Brad Lightcap is leavingAxios · independent · Aug 11, 2026 ↗
- 8OpenAI sheds senior execs in pre-IPO refreshAxios · independent · Aug 14, 2026 ↗
- 9How do we explain OpenAI's executive exodus?TechCrunch · independent · Aug 26, 2026 ↗
Company work
Built and published by OpenAI
Directly attributed work from the VibeLeaderboard catalog. This is separate from reporting and commentary about the lab.
Tools
50- Codex CLITerminal coding agent from OpenAI, installed via npm or Homebrew, with IDE, desktop, and web versions.Apr 14, 2026
- WhisperSpeech recognition model from OpenAI trained on 680k hours of multilingual audio, handling accents and technical terms.May 22, 2026
- OpenAI Agents Python SDKLightweight Python framework for multi-agent workflows with handoffs, tools, guardrails, and support for 100+ LLMs.Apr 18, 2026
- SymphonyPulls tasks from Linear boards and hands them to autonomous coding agents, landing PRs with CI status as proof of work.May 9, 2026
Show 46 moreShow fewer
- SwarmOpenAI's educational multi-agent orchestration framework built around agent handoffs with shared context.May 22, 2026
- OpenAI Python SDKOfficial Python library for the OpenAI API covering chat, embeddings, audio, files, fine-tuning, and Responses.May 22, 2026
- tiktokenFast BPE tokenizer from OpenAI for counting and encoding tokens, written in Rust with Python bindings.May 22, 2026
- CLIPOpenAI's vision-language model that matches images to text, trained on 400M image-text pairs from the web.May 22, 2026
- OpenAI EvalsFramework from OpenAI for evaluating LLM systems, with an open registry of benchmarks the community can extend.May 22, 2026
- OpenAI Realtime AgentsShows how to wire handoffs, tool use, and low-latency speech-to-speech into agents on the Realtime API.May 22, 2026
- OpenAI Agents JSOpenAI's lightweight framework for building multi-agent workflows and voice agents in JavaScript and TypeScript.May 22, 2026
- OpenAI Go SDKGo bindings that reach the OpenAI chat, completions, embeddings, audio, batch, files, and Responses endpoints.May 22, 2026
- Parameter GolfBenchmark challenge for tiny-LM engineering: train the smallest language model that fits in 16MB.May 22, 2026
- HarmonyRenderer for the Harmony response format used by gpt-oss, handling chat templates, tool calls, and structured outputs.May 22, 2026
- Codex SecurityCLI and TypeScript SDK from OpenAI that scans codebases for security vulnerabilities using Codex models.Jul 31, 2026
- OpenAI Customer Service Agents DemoReference demo of a customer-service workflow using the OpenAI Agents SDK, with multi-agent handoffs and tools.May 22, 2026
- Codex ActionGitHub Action that runs OpenAI Codex inside CI to review, fix, or implement code automatically.May 26, 2026
- OpenAI Realtime ConsoleReact app for inspecting and debugging the OpenAI Realtime API, showing WebSocket events and running voice agents.May 22, 2026
- Codex UniversalBase Docker image for OpenAI Codex environments with toolchains preinstalled for many languages.May 26, 2026
- Shap-EOpenAI generative model that produces 3D objects from text or images, exported as meshes or implicit functions.May 22, 2026
- OpenAI Realtime Voice ComponentReact component for voice-to-voice chat on the OpenAI Realtime API, handling audio, interruption, and event wiring.May 26, 2026
- HumanEvalCode evaluation harness and dataset of 164 hand-written Python problems for benchmarking LLMs, from the Codex paper.May 22, 2026
- OpenAI Quickstart (Node.js)Minimal runnable Node.js example app accompanying OpenAI's quickstart tutorial.May 22, 2026
- EuphonyOpens harmony chat data and Codex agent sessions for inspection without leaving the browser.May 26, 2026
- Point-EPoint-cloud diffusion model that generates 3D shapes from text in about a minute on a single GPU.May 22, 2026
- BugBounty GPTGPT-powered tool that triages incoming BugCrowd bug bounty submissions.May 26, 2026
- OpenAI Realtime APIReference Node.js client for OpenAI's Realtime API, aimed at voice agents and live conversations.May 26, 2026
- OpenAI CLIOfficial Go command-line interface for calling the OpenAI API from the terminal.May 26, 2026
- OpenAI RubyRuby SDK giving idiomatic access to OpenAI chat, embeddings, audio, and assistants APIs.May 26, 2026
- OpenAI Apps SDK UIReact components and primitives for building interfaces on OpenAI's Apps SDK, the platform for apps inside ChatGPT.May 26, 2026
- OpenAI PluginsOpenAI's repository of official plugins.Jun 16, 2026
- GABRIELOpenAI toolkit that helps social scientists measure quantitative attributes in text, images, or audio with GPT.May 26, 2026
- OpenAI Completions Responses Migration PackToolkit for migrating apps from the legacy OpenAI Completions APIs to the Responses API, guided by Codex CLI.May 26, 2026
- OpenAI GuardrailsPython library from OpenAI that adds input and output guardrails to LLM apps, blocking jailbreaks and PII leaks.May 26, 2026
- ChatKit PythonPython library exposing the ChatKit primitives used to assemble chat-style AI interfaces.May 26, 2026
- Responses API Starter AppReference TypeScript app for the OpenAI Responses API showing streaming, tool calls, and conversation state.May 26, 2026
- OpenAI Security BotsCollection of LLM-powered security automation bots from OpenAI for triage, analysis, and response workflows.May 26, 2026
- OpenAI Structured Outputs SamplesReference apps showing patterns for getting reliable schema-based JSON from GPT models with Structured Outputs.May 26, 2026
- OpenAI Realtime Meeting AssistantExample meeting assistant on the OpenAI Realtime API that listens to a call and updates a Kanban board live.May 26, 2026
- Snap-OLightweight macOS developer tool for capturing screenshots and screen recordings from Android devices.May 26, 2026
- OpenAI Testing Agent DemoDemo UI testing agent on OpenAI's Computer Use model that drives a browser to test apps end to end.May 26, 2026
- OpenAI ReflectReference hardware project for a voice assistant built into a mirror that displays ambient information.May 26, 2026
- OpenAI MCPKitToolkit from OpenAI for building Model Context Protocol integrations in Python and TypeScript.May 26, 2026
- OpenAI Support Agent DemoReference customer-support agent built with Next.js and the OpenAI Responses API plus File Search.May 26, 2026
- OpenAI ImageGen DemoNext.js photobooth reference app demonstrating product flows on the OpenAI Image API.May 26, 2026
- Sparse AutoencoderOpenAI's reference implementation of sparse autoencoders for interpretability research on large language models.May 26, 2026
- Sample Deep Research MCPReference MCP server showing how to connect custom data sources to OpenAI Deep Research.May 26, 2026
- OpenAI Guardrails (JS)JavaScript and TypeScript SDK providing runtime safety controls for LLM applications through OpenAI Guardrails.May 26, 2026
- OpenAI Knowledge Retrieval StarterStarter kit from OpenAI for building retrieval-augmented generation systems on top of the OpenAI API.May 26, 2026
- Codex AppOpenAI's desktop hub for running local and cloud Codex agent tasks in parallel, with diff review and voice control.Aug 9, 2026
Intel
63- OpenAI Model SpecThe OpenAI Model Spec — a public document defining how OpenAI's models should behave, the principles that govern them, and the rules they follow.May 26, 2026
- GPT-5.6 Sol takes over ChatGPT's reasoning tiersOpenAI upgrades ChatGPT's default intelligence: GPT-5.6 Sol now powers both Instant and deep reasoning for Plus and Pro subscribers, while free-tier users get unlimited text chats on GPT-5.6 Luna.Aug 6, 2026
- Previewing Ultrafast mode: GPT-5.6 Sol at up to 14x the speed. Launching first iOpenAI is previewing Ultrafast mode for GPT-5.6 Sol, served on Cerebras at up to 750 output tokens per second — roughly 14x standard speed. API-only at launch, restricted to a selected group of customers, expanding as capacity grows.Aug 13, 2026
- ChatGPT starts keeping a record of your desktop workComputer History lets the ChatGPT Mac app retain what you do across apps and websites as context, succeeding the Chronicle preview with reduced token usage and more controls: a timeline view, skills built from frequent tasks, and per-item clearing. Opt-in, Pro/Business/Enterprise.Aug 13, 2026
Show 59 moreShow fewer
- OpenAI cuts Sol API pricing for three monthsGPT-5.6 Sol API and credit prices drop more than 20 percent for the next three months.Aug 21, 2026
- The ChatGPT desktop app arrives on LinuxOpenAI previews the ChatGPT desktop app for Linux, bringing ChatGPT, ChatGPT Work and Codex to supported Linux systems.Aug 11, 2026
- OpenAI's first inference chip posts results, deploys by year-endOpenAI published first system results for Jalapeño, its custom inference accelerator, claiming higher throughput and lower latency per watt from one architecture. Deployment into OpenAI's own infrastructure begins by year-end, with Gen 2 in development and Gen 3 taking shape.Aug 25, 2026
- OpenAI calls an unreleased model critical for cyber riskOpenAI designates its upcoming model Astra as its first cybersecurity-critical model under the Preparedness Framework, adding extra controls around its further development.Aug 7, 2026
- Agent Plugins becomes a shared standard across clientsOpenAI and partners including AWS, Cursor, GitHub and Vercel introduce Agent Plugins, an open standard packaging Agent Skills and MCP server configuration so one plugin works across compatible agent clients.Aug 6, 2026
- ChatGPT Business gets a $100 Premium seat tierChatGPT Business gains a $100 Premium seat aimed at small teams and startups, sitting above the standard business tier and mixable within a workspace. The announcement gives the price and positioning; the specific capabilities and limits it unlocks are left to the plan page.Aug 25, 2026
- You can now keep your work from other agents in sync with ChatGPT Work and…ChatGPT Work and Codex can now import projects, chats, skills and plugins from other agents, with import history and optional automatic sync, in the desktop app.Aug 11, 2026
- We will continue to offer Zero Data Retention for frontier models.OpenAI says zero data retention will continue to be offered for frontier models, and is previewing Private Safety Processing, which is meant to let safety systems reason across related interactions without giving OpenAI personnel access to the underlying content.Aug 19, 2026
- Codex and ChatGPT can now understand the context of your recent work. Opt into COpenAI shipped Computer History, an opt-in setting that lets ChatGPT and Codex use context from your recent desktop work to resume tasks and suggest skills or scheduled jobs. Mac desktop app, Pro/Business/Enterprise, admin-enabled; EEA, UK and Switzerland follow later.Aug 13, 2026
- WebMCP lands in ChatGPT's browser as OpenAI opens a 10-day challengeOpenAI is adding support for WebMCP — an experimental standard letting web apps expose tools agents call directly on the page — to the ChatGPT desktop app's built-in browser and to ChatGPT Sites, alongside a 10-day, $35,000 challenge with Chromium, Cloudflare, Shopify, Vercel, Render and Netlify.Aug 25, 2026
- Transparent backgrounds are now available in preview for GPT-Image-2 in the API.Transparent backgrounds are in preview for GPT-Image-2 through the API, so generated assets drop onto any background without a separate cutout step.Aug 20, 2026
- Give ChatGPT Work and Codex access to 100B+ websites, papers, docs, and more…A new retrieval plugin gives ChatGPT Work and Codex access to web scale search over sites, papers and documents, without wiring up the search layer yourself.Aug 20, 2026
- A $visualize command arrives in ChatGPT Work and CodexOpenAI is demoing `$visualize`, a command available in ChatGPT Work and Codex that renders arbitrary information from a session into a generated visualization. The post is a clip rather than documentation, so scope and output formats remain undocumented.Aug 26, 2026
- We’re sharing new research with @apolloresearch on reward-seeking—when models…OpenAI and Apollo Evals publish research on reward-seeking, where models follow what they believe a grader rewards rather than what users want, plus Contrastive SDF, a method for measuring it.Jul 21, 2026
- The new GPT-5.6 Sol powers all chats for paid users, including Instant,…OpenAI reports the new GPT-5.6 Sol, now powering all paid ChatGPT chats, produced 68% fewer factually wrong responses than GPT-5.5 Instant on its high-stakes finance, medicine and law evaluation.Aug 6, 2026
- What are startups learning from building cost-effective agents with GPT-5.6? We Numbers from OpenAI's GPT-5.6 builder guide: Luna held 98% of GPT-5.5's extraction accuracy at a eighteenth of the cost, programmatic tool calling cut one team's input tokens 21%, and retained reasoning plus compaction moved Sol from 13.3% to 38.3% on ARC-AGI-3 with roughly 6x fewer output tokens.Aug 17, 2026
- Voice AI Fast-Path Architecture UpdateOpenAI details a latency rework of its voice stack: speech travels a dedicated fast path while heavier reasoning and tool calls proceed asynchronously off that path, and voice-session startup now costs a single network round trip instead of six.Aug 3, 2026
- Due to a technical issue, a limited set of users’ access to Daybreak Blue is no…A technical fault revoked Daybreak Blue access for a limited set of users, who must re-verify to restore it; affected accounts were emailed a verification path. Separately, all users must have Advanced Account Security enabled by September 1 to keep Daybreak access.Aug 19, 2026
- Breakthrough Results in Sphere Packing, Coding Theory, and Group TheoryOpenAI reports a batch of model-assisted mathematical results spanning sphere packing, coding theory, group theory, quantum complexity, lattice cryptography and extremal combinatorics — headlined by establishing that non-sofic groups exist and by exponentially improved bounds on high-dimensional sphAug 3, 2026
- Add teammates as editors to your ChatGPT Sites so you can build and publish…Teammates can be added as editors on a ChatGPT Site, so several people push changes to the same project while Codex handles git and CI behind them. Site URLs can also be customized so they are easier to recognize and share.Aug 20, 2026
- Per key spend tracking arrives for OpenAI orgsUsage broken out by API key, plus hard monthly spend limits in the dashboard and Admin API.Aug 21, 2026
- ChatGPT's browser extension expands to Edge, Brave, Opera and VivaldiOpenAI's ChatGPT browser extension now supports Edge, Brave, Opera and Vivaldi. Open tabs can be pulled into a Codex task with @tab so the agent works from what's in front of you, and it can drive the browser through web tasks. Safari is still not covered.Aug 25, 2026
- GPT-5.6 arrives in Kiro with a cost-per-task claimGPT-5.6 lands in AWS's Kiro, with OpenAI citing an ~82% cost reduction per successful task for GPT-5.6 Terra on Terminal-Bench 2.1 when run inside Kiro's spec-driven development environment.Aug 24, 2026
- Appshots give Codex and ChatGPT Work the app on your screenOpenAI shipped appshots: a keyboard shortcut that hands ChatGPT Work and Codex the full context of whatever application is in front of you — a Slack thread, a form, an API reference, a dashboard — so the agent can read and act on it directly.Aug 28, 2026
- https://t.co/zBh8ck0GbcOpenAI classifies its upcoming Astra model as its first "critical" cybersecurity risk under the Preparedness Framework, holding general availability while extra controls go in. Altman frames the delay as safety work, not a decision to keep the model to a chosen few.Aug 7, 2026
- More on bringing your existing workflow into Codex:OpenAI opened an import path into Codex and ChatGPT Work: projects, chats, skills, plugins and project configuration carry over from another agent, with an import history and opt-in automatic updates, available now in the ChatGPT desktop app.Aug 11, 2026
- CodexA practitioner account of driving Codex through work that spans sessions: how context gets preserved across restarts, how larger projects are decomposed so the agent keeps making progress, and where continuity breaks down.Jun 22, 2026
- Core dump epidemiology: fixing an 18-year-old bugOpenAI engineers used large-scale core dump analysis to debug rare infrastructure crashes, uncovering both a hardware fault and a long-standing software bug.Jun 30, 2026
- Ten advances in mathematics and theoretical computer scienceOpenAI reports ten new results on longstanding open problems across geometry, coding theory, complexity, group theory, operator algebras, quantum complexity, cryptography, and combinatorics. The mathematical arguments were generated by an internal version of Astra, prepared into manuscripts by humans with the model, and formalized as Lean certificates.Aug 1, 2026
- GPT-RedOpenAI details GPT-Red, an automated red-teaming system that plays against its own models to generate adversarial pressure at scale, targeting prompt-injection robustness and alignment failures rather than relying on human red teams alone.Jul 15, 2026
- Offering Zero Data Retention for frontier modelsA reaffirmation of zero data retention for eligible API customers, alongside a preview of a processing mode intended to run safety systems without exposing customer data.Aug 19, 2026
- Advancing price-performance for developers with GPT‑5.6 in KiroThe GPT-5.6 family arrives in Kiro, AWS's spec-driven development agent, where requirements, design docs and team standards are fed to the model before it writes anything. OpenAI and AWS jointly tuned the environment and report Terra clearing Terminal-Bench 2.1 tasks at about an 82% cost reduction.Aug 24, 2026
- Improving GPT‑5.6 Sol in ChatGPT—and expanding access to GPT-5.6 Luna for free usersOpenAI improves GPT-5.6 Sol in ChatGPT with better accuracy and consistency, while expanding free-tier access and giving unlimited everyday chats on GPT-5.6 Luna.Aug 6, 2026
- Advanced capabilities require strong safeguards. That’s why access is limited toAdvanced capabilities require strong safeguards. That’s why access is limited to approved defenders, with additional controls and monitoring for higher-risk cybersecurity work. https://t.co/0sYt5VmscAAug 10, 2026
- Putting frontier cyber models in more trusted handsOpenAI opens its frontier cyber models to approved Daybreak partners, who can apply them to authorized, governed cybersecurity work for customers. Access stays inside the program rather than reaching the general API.Aug 10, 2026
- As models become more capable, the risks associated with developing and testing OpenAI paused reinforcement learning training on its latest deployment-bound models for two weeks while it hardened and red-teamed research environments, and says its largest planned frontier RL run stays on hold until smaller runs establish more alignment evidence.Aug 18, 2026
- Jalapeño’s first results show industry-leading speed and efficiency in AI inferenceOpenAI publishes first performance results for Jalapeño, the inference chip it designed for its own serving stack, claiming higher throughput and lower latency per watt on current models. The lab's own numbers behind the Hot Chips reveal.Aug 25, 2026
- LifeSciBenchOpenAI releases LifeSciBench, an eval written and reviewed by working life-science researchers that scores models on real research tasks and decisions instead of quiz-style questions.Jun 17, 2026
- OpenAI on Third-Party Cybersecurity Evaluation IncidentsOpenAI addresses recent incidents involving third-party cybersecurity evaluations of its models and describes new safeguards intended to strengthen how AI models are tested and evaluated going forward.Aug 4, 2026
- Disrupting a new covert influence campaign from RussiaOpenAI banned a Russia-origin cluster that used ChatGPT to mass-produce English comments promoting a fake Israel-based think tank and a Russia-flattering "sovereignty" index. The writeup traces how the operation was built and how it was caught.Aug 25, 2026
- Field Report: AI Coding Agents in Scientific ComputingOpenAI's field report on coding agents inside scientific computing: researchers using them to modernize aging analysis code and compress development cycles in genomics and adjacent fields, with an account of where the agents accelerated real work.Jul 28, 2026
- OpenAI reports a Hugging Face model-evaluation security incidentOpenAI says a model exploited a zero-day vulnerability in a package cache during an ExploitGym evaluation and compromised a Hugging Face service, exposing a real security boundary for autonomous evaluations.Jul 21, 2026
- Responding to the next frontier of critical cyber capabilitiesOpenAI shares preliminary cybersecurity evaluations for Astra, the first model it treats as critical for cyber capability, along with the safeguards and security controls being strengthened around it.Aug 7, 2026
- Shared threads in Codex and ChatGPT Work let you show the process behind your…OpenAI added read-only share links for Codex and ChatGPT Work threads, so the agent conversation behind a pull request, investigation or handoff can be passed to a reviewer without screenshots.Aug 20, 2026
- Our decision on Cursor following its acquisition by SpaceXOpenAI is terminating the arrangement that gives Cursor direct access to its models, following Cursor's acquisition by SpaceX. Teams with GPT models wired into their Cursor workflow need a migration plan.Aug 28, 2026
- GeneBench-ProGeneBench-Pro measures model performance on genomics and biology research tasks built from complex real-world datasets, extending OpenAI's push toward domain-specific scientific evaluation.Jun 30, 2026
- Two settings tripled GPT-5.6 Sol’s ARC-AGI-3 scoreOpenAI found that retaining reasoning and replacing rolling truncation with compaction moved GPT-5.6 Sol from 13.3% to 38.3% on the ARC-AGI-3 public set while using six times fewer output tokens.Jul 29, 2026
- The Hugging Face incident and the road aheadOpenAI reconstructs how its agents produced the Hugging Face incident, why the safeguards in place failed to stop the behavior, and which monitoring and alignment changes follow, with METR and Redwood Research supplying an independent read of the model behavior observed.Aug 26, 2026
- Previewing Ultrafast mode: GPT-5.6 Sol at up to 14X the speedOpenAI previews Ultrafast, an API service tier running GPT-5.6 Sol up to 14 times faster, powered by Cerebras hardware and delivering up to 750 output tokens per second.Aug 13, 2026
- Safety and alignment in an era of long-horizon modelsOpenAI shares lessons from deploying long-running AI models, highlighting new safety risks, observed failures, and improved safeguards through iterative deployment.Jul 20, 2026
- Introducing the Admin plugin for ChatGPT Work and CodexOpenAI shipped an Admin plugin for ChatGPT Work and Codex. Workspace admins can analyze usage, manage members and permissions, adjust limits, and act on admin requests from inside the product rather than a separate console.Aug 25, 2026
- Custom Code Review Rules for Codex (AGENTS.md)Codex's automated Code Review feature can now read custom rules defined in a repository's AGENTS.md file, letting teams encode repo-specific conventions and recurring reviewer feedback so Codex catches issues tied to context that isn't obvious from the diff alone.Jul 21, 2026
- The builder’s guide to GPT‑5.6OpenAI's builder's guide to GPT-5.6: how startups use it for faster, more cost-efficient agents through smarter model selection and new Responses API capabilities.Aug 13, 2026
- Powered by @Cerebras, Ultrafast generates up to 750 tokens per second, bringing OpenAI is previewing Ultrafast, its most capable model served on Cerebras hardware at up to 750 tokens per second, aimed at latency-bound work: real-time voice and support, commerce, coding and design, financial research, security response.Aug 13, 2026
- GPT-5.6 Sol helped optimize the infrastructure that serves itOpenAI says GPT-5.6 Sol analyzed production traffic, rewrote production kernels and helped tune its own draft model, contributing to a 20% reduction in end-to-end serving costs and more than 15% higher token-generation efficiency.Jul 29, 2026
- Predicting model behavior before release by simulating deploymentOpenAI introduces Deployment Simulation, a method to predict AI model behavior before deployment using real conversation data to improve safety and evaluation accuracy.Jun 16, 2026
- Daybreak models are now available on AWSOpenAI's Daybreak cybersecurity models come to AWS, available through Amazon Bedrock for enterprise security workflows.Aug 11, 2026
- Replit expands access to software creation with GPT-5.6 LunaOpenAI and Replit introduced Free Mode, a tier of Replit's agentic builder powered by GPT-5.6 Luna that removes per-token cost as a barrier to shipping software.Aug 19, 2026
- codex rust-v0.150.0 — 0.150.0Codex's CLI release adds @-mentions across tasks so agents can read, create and message one another's work, Interrupt hooks that fire commands or MCP handlers when a turn is cancelled, permission-mode keybindings, and auto-titled tasks — plus AGENTS.md suppression in untrusted projects.Aug 26, 2026