Artificial Intelligence

OpenAI Announces GPT-5.3-Codex

Is GPT-5.3-Codex a revolutionary leap in AI capabilities or an impractical tool with usability flaws?

Published FEB 6

Story

Above: The OpenAI logo displayed on a smartphone. Image credit: Samuel Boivin/NurPhoto/Getty Images

The Facts

OpenAI released its new agentic coding model, GPT-5.3-Codex, on Thursday, which the AI company claims is its "most capable" to date, combining GPT‑5.2-Codex's frontier coding with GPT‑5.2's reasoning and professional knowledge.
Openai
Mashable
VentureBeat
The new model is 25% faster than its previous iterations and substantially more powerful, with GPT-5.3-Codex achieving scores of 56.8% on SWE-Bench Pro, 77.3% on Terminal-Bench 2.0 and 64.7% on OSWorld-Verified, compared to GPT-5.2-Codex's respective scores of 56%, 64% and 38.2%.
Openai
According to OpenAI, GPT-5.3-Codex was also "instrumental" in its own creation, with the codex team using early versions of the model to debug and analyze its training, propose fixes and support its launch by dynamically scaling GPU clusters to maintain stable latency and adjust to traffic surges.
Openai
The codex's launch is the first classed as "High capability" for cybersecurity under OpenAI's Preparedness Framework. In accordance with this designation, OpenAI has activated multiple associated safeguards — including model safety training, conversation monitoring, trust-based access and actor-level enforcement strategies.
Openai
Despite its capabilities, OpenAI stated that GPT-5.3-Codex failed to meet the "High capability" performance criteria for AI self-improvement, which requires a model to be capable of real-world software or research engineering, as well as machine learning debugging and diagnostics.
Openai
The release of GPT-5.3-Codex coincided with Anthropic's announcement of Claude Opus 4.6, an upgrade that the rival AI company claimed made Claude more careful and reliable, more capable of sustaining agentic tasks for longer and better at identifying its own errors.
Anthropic

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Narrative A

The launch of GPT-5.3-Codex represents a monumental leap in AI capabilities, with the model achieving unprecedented performance across every major benchmark, all while operating at faster speeds and using fewer tokens. In application, the codex holds the promise to revolutionize the working environment, with the tool capable of performing complex tasks across multiple domains for maximum efficiency.

Openai

Sam Altman on X

Narrative B

While GPT-5.3-Codex is a substantial upgrade, OpenAI has made numerous adjustments that limit its functional capabilities. The codex's slow running time, for example, makes the model impractical for many real-world workflows, where developers need rapid iteration cycles. At the same time, its status updates can be unreliable and its run summaries too technical, which could alienate less experienced developers.