OpenAI says GPT-5.3-Codex goes beyond an agent that can code “to an agent that can do nearly anything developers and professionals can do on a computer”
Expanding Codex across the full spectrum of professional work on a computer. — Join the Codex app waitlistForums:Hacker Newsandr/singularityForums:Hacker News:GPT-5.3-Codexr/singularity:OpenAI released GPT 5.3 Codex
OpenAI
Context & Ripple Effects
Codex began as a software-engineering agent for ChatGPT business tiers, then GPT-5-Codex was positioned around more adaptive reasoning for agentic coding. Subsequent work on long-horizon code changes extended that trajectory before this broader professional-work claim.
The significance is not simply a new coding model: OpenAI is positioning Codex as a computer-work agent and opening an app waitlist, potentially making the product’s interface as important as its underlying model.
First-order effects
OpenAI expands Codex’s stated remit from software engineering to tasks across developers’ and professionals’ computer work, while using the app waitlist to control initial access.
Existing Codex users must evaluate it as a broader workflow tool rather than only a coding assistant, building on its earlier rollout to ChatGPT business users.
Second-order effects
Developer-tool and workplace-software vendors face pressure to make their agents operate across more applications and multi-step workflows, rather than confining them to code generation or single features.
Organizations testing Codex will need to decide which computer actions can be delegated and where review remains necessary, increasing the importance of workflow-level permissions and oversight.
Third-order effects
If these broader agents prove dependable, the competitive unit shifts from a coding copilot to an agentic work surface that coordinates tasks across desktop and browser tools.
That shift could favor platforms able to combine capable models, persistent task context, and governed access to workplace systems; the breadth of OpenAI’s claim still requires real-world validation.
The trend: This is one data point in workplace-agent generalization, in which coding agents are being recast as general operators of professional software workflows.
GPT-5.3-Codex — Cheaper (fewer tokens, higher intelligence) — you can interrupt it — big focus on coding — first LLM to be used extensively in its own development — openai.com/index/introd... [image]
@sama
Sam Altman on x
This is our first model that hits “high” for cybersecurity on our preparedness framework. We are piloting a Trusted Access framework, and committing $10 million in API credits to accelerate cyber defense. https://openai.com/...
@sama
Sam Altman on x
I love building with this model; it feels like more of a step forward than the benchmarks suggest. Also you can choose “pragmatic” or “friendly” for its personality; people have strong preferences one way or the other!
@zephyr_z9
@zephyr_z9 on x
Need more speed ups like this to enable long horizon tasks and not bore the user to death
@youjiacheng
You Jiacheng on x
5.3 only uses 48% tokens of 5.2 (both xhigh) and has 25% higher TPS. the wallclock speedup is +160% (2.6× speed), crazy.
@nvidianewsroom
@nvidianewsroom on x
OpenAI's new GPT-5.3-Codex was co-designed for, trained with, and is served on NVIDIA GB200 NVL72 systems, accelerating the frontier of AI. We're excited to see what developers build next with this breakthrough in code intelligence.
@scaling01
@scaling01 on x
GPT-5.3 Codex absolutely demolished Opus 4.6 (65.4%) on Terminal Bench 2 just minutes after its launch! [image]
@scaling01
@scaling01 on x
OpenAI released GPT-5.3-Codex with massive improvements in reasoning efficiency! [image]
@gdb
Greg Brockman on x
gpt-5.3-codex — smarter, faster, and very capable at tasks like making presentations, spreadsheets, and other work products. Codex becoming an agent that can do nearly anything developers and professionals can do on a computer. [image]
@sama
Sam Altman on x
It was amazing to watch how much faster we were able to ship 5.3-Codex by using 5.3-Codex, and fore sure this is a sign of things to come.
@marvinvonhagen
Marvin von Hagen on x
“GPT-5.3-Codex is our first model that was instrumental in creating itself.”
@natolambert
Nathan Lambert on x
“Due to GPT-5.3-Codex being so different from its predecessors, the data from alpha testing exhibited numerous unusual and counter-intuitive results” Sounds worth giving a go. Big changes are good.
@sama
Sam Altman on x
GPT-5.3-Codex is here! *Best coding performance (57% SWE-Bench Pro, 76% TerminalBench 2.0, 64% OSWorld). *Mid-task steerability and live updates during tasks. *Faster! Less than half the tokens of 5.2-Codex for same tasks, and >25% faster per token! *Good computer use.
@openaidevs
@openaidevs on x
GPT-5.3-Codex is our first model that was instrumental in creating itself. The Codex team used early versions to debug training, manage deployment, and diagnose test results and evaluations, accelerating its own development.
@kimmonismus
@kimmonismus on x
GPT-5.3 codex!! What a day!!
@synthwavedd
Leo on x
THEY RELEASED 5.3 CODEX THE SECOND OPUS 4.6 DROPPED LMAO [image]
@danielkarunaa
Daniel on x
aaaaand gpt-5.3-codex in codex cli now! great timing (v0.98.0) [image]
@sarahwooders
Sarah Wooders on x
Not available on the API in the name of “safety” (classic OpenAI), so impossible to properly evaluate or try out I'll be sticking to Opus 4.6 [image]