We’ll be livestreaming the OpenAI DevDay keynote 2026-09-29T17:00:00Z. We’d love to have you join us!
In the calendar. Locked and loaded. Ready to stream all the way in chilly Sweden!
So glad it’s at 7 P.M. my time. 
See Dys I told you. This forum just runs a little slower 
See you there! from NYC <3
Devday Bingo - my lucky square
OPENAI REINTRODUCES GPT-4.5 AT DEVDAY 2026 FOLLOWING RE-EVALUATION OF COMPUTE REQUIREMENTS AND COSTS
Large, general-purpose GPT model returns for developers with its original model characteristics
SAN FRANCISCO, Sept. 29, 2026 — OpenAI today announces at DevDay 2026 the re-introduction of GPT-4.5, bringing the model back to developers following a re-evaluation of the computation required to serve it and the associated costs of operating large-scale, non-reasoning models. GPT-4.5 returns with the core specifications of the original model: a 128,000-token context window, up to 16,384 output tokens, an October 1, 2023 knowledge cutoff, text input and output and image input, together with support for function calling, Structured Outputs, streaming and system messages. Originally introduced as gpt-4.5-preview, GPT-4.5 represents an approach centered on scaling pre-training and unsupervised learning rather than inference-time reasoning. The model is designed for broad world knowledge, natural interaction, instruction following, creativity, writing, communication, programming and agentic planning. At re-launch, API pricing is $75 per million input tokens, $7.50 per million cached input tokens and $150 per million output tokens, placing it under the pricing of gpt-5.4-pro also served on the API, with low-latency answering.
OpenAI first introduces GPT-4.5 as a research preview on February 27, 2025, describing it at the time as its largest and most knowledgeable GPT model and making it available to developers on paid API tiers as well as initially to ChatGPT Pro users. Its scale also makes it unusually compute-intensive, and OpenAI states at launch that it is evaluating whether the model can be supported economically over the long term. That evaluation leads to an unusually short initial API lifecycle: OpenAI announces the model’s deprecation on April 14, 2025 and ends API access on July 14, 2025, less than five months after introduction, while GPT-4.5 remained available in ChatGPT Pro until June 27, 2026. Improvements in infrastructure, capacity planning and the economics of serving large models now allow OpenAI to revisit that decision. Re-introducing GPT-4.5 gives developers access once again to a model with a distinct profile alongside newer generations of reasoning and general-purpose systems, while preserving the characteristics that made the original release useful for applications emphasizing nuance, creativity and natural interaction.
About OpenAI
OpenAI is an AI research and deployment company. Our mission is to ensure that artificial general intelligence benefits all of humanity. OpenAI develops models, products and developer platforms that enable people and organizations to build with increasingly capable artificial intelligence systems.
on that note:
I’m watching it now, but it’s sad, there’s no a side chat there (on YT), but think, it is for a reason.
Welcome back, @luxana.ao … feel free to discuss here. We’ve got a great community, imho!
Thank you, I’m watching, listening… and playing Warcraft Forever Beta (testing) at the same time. I like the React part, I was learning React last year at my uni, I hope I’ll have more AI this tear as I’m mainly interested in AI & Ethics, not only in coding. And when I’m finishing this reply, Astra possibilities are shown now 
Off top question, is there a Discord server for this forum?
sad, the YT video was short, but was nice to watch it.
I think this year sets the record for the most things announced in the shortest amount of time:
- Dots: persistent personal agents that handle ongoing work.
- Dots roadmap: SMS support coming soon; teams of dots planned.
- ChatGPT Space: shared workspace for people and agents.
- Pages: collaborative documents with research, images, charts, and agent contributions.
- Collaborative presentations: shared AI-assisted slide creation, coming soon.
- Specialist enterprise dots: preview of company-configured agents for functions such as accounting, marketing, and legal.
- Microsoft Agent 365 integration: planned management integration for specialist dots.
- GPT-6.1 Sol: new model positioned near Astra’s capabilities at lower token prices.
- Ultrafast: premium inference speed; Astra available immediately, Sol coming soon.
- Pro 500: new subscription with 25× the Plus allowance and Ultrafast access.
- Pro 200 reopened: subscriptions available again.
- Decisions API: preview for fast classification, routing, and choosing agent actions.
- Codex in the cloud: tasks continue with your laptop closed, accessible across devices.
- Codex Security Cloud: vulnerability scanning, investigation, deduplication, and proposed fixes.
- Agents API public beta: managed agent infrastructure with memory and multi-agent capabilities.
- Computer use in Agents API: agents can interact directly with software.
- Bedrock Managed Agents powered by OpenAI: agents running within AWS.
- Private Intelligence: Private Safety Processing and a Private Inference preview.
- API performance improvements: reported reductions in initial response latency and tool/workflow execution time.
- Refreshed Codex CLI: including two-way voice interaction.
- Sign in with ChatGPT: eligible subscription usage in third-party products, starting with 16 partners.
- Plugin extensions: richer applications, editors, dashboards, and workspaces inside ChatGPT and Codex.
- Meetings plugin: calendar-connected notes and collaborative follow-up work.
- Partner extensions: demonstrations included Figma and Adobe Photoshop.
- Plugins in Sites: connected applications and personalized data within shared Sites.
- Plugin discovery during conversations: relevant plugins suggested as you work.
- Improved plugin submissions: review tracking, clearer feedback, human review requests, and simpler updates.
- OpenAI Marketplace: enterprise commitments can fund eligible partner products.
- Open-source models through Baseten: available through Marketplace.
- OpenClaw enterprise harness: briefly announced, with details reserved for a later session.
- Worldwide usage/rate-limit reset: announced and triggered during the closing segment.
The official recap also lists these additions, which weren’t mentioned during the livestream:
- Redesigned Code Review, including automated cloud reviews.
- Additional CLI improvements:
/agents, prompt editing, session resumption, and worktrees. - Plugin Creator improvements and better recommendations.
- MCP events to trigger plugin automations.
- Teams and shared recurring tasks.
- @ChatGPT in Slack and Microsoft Teams.
- Shareable profiles showcasing Sites and plugins.
The computer use announcement particularly caught my attention. I’m a developer from Yemen exploring an idea to take this capability further: a stronger bridge between AI systems and real-world desktop applications.
Rather than building another AI model, I want to develop a reusable knowledge and skills layer that helps existing agents understand applications and their workflows—not just where to click.
The idea combines learning from demonstrated workflows with a structured knowledge base covering widely used desktop software, especially accounting and ERP systems, including legacy applications. This knowledge would be organized by application, country, and region, making it relevant to the software and business practices people actually use in different markets.
Instead of repeatedly rediscovering how an application works, agents could draw on application-specific knowledge and reusable skills. The goal is more specialized, locally relevant assistance, with fewer tokens and faster, more reliable execution.
Would anyone here be interested in supporting or collaborating on an early-stage project in this direction? I’d love to connect with technical collaborators, mentors, or potential backers who share this vision.
A humble developer from Yemen, with an ambitious idea.
That’s gonna be a cool multiple years long ride. Enjoy!