GPT-5.3
- Provider
- OpenAI
- Status
- Superseded
- Context
- 400,000 tok
GPT-5.3 is the OpenAI model generation that sat between GPT-5.2 (December 2025) and GPT-5.4 (March 2026) in the fast-moving GPT-5 line. Its most significant member is GPT-5.3-Codex, released on 24 February 2026 — OpenAI’s agentic coding model, which the company described as achieving state-of-the-art results on SWE-bench Pro at launch, with strong Terminal-Bench 2.0 and OSWorld-Verified scores, roughly 25% faster and more token-efficient than prior Codex models. The general GPT-5.3 model powered ChatGPT’s “GPT-5.3 Instant” default until GPT-5.5 Instant replaced it on 5 May 2026.
GPT-5.3 had a short life as a frontline model: GPT-5.4 arrived within two weeks and GPT-5.5 a few weeks after that. It is now superseded, though GPT-5.3-Codex remains available via the API and is the part of the line most worth knowing about.
Quick specs
| Provider | OpenAI |
| Tier | Strong (superseded) |
| Released | 24 February 2026 (GPT-5.3-Codex) |
| Status | Superseded by GPT-5.4 / GPT-5.5 |
| Standout variant | GPT-5.3-Codex (agentic coding) |
| API model ID | gpt-5.3-codex |
| Context window | 400,000 tokens |
| Max output | 128,000 tokens |
| GPT-5.3-Codex price | $1.75 / $14.00 per MTok |
| Coding | SOTA on SWE-bench Pro at its February 2026 release (per OpenAI) |
| Best for | Agentic coding (via GPT-5.3-Codex) |
| Limitations | Superseded within weeks; general-model benchmarks not separately published |
What GPT-5.3 is
GPT-5.3 is part of OpenAI’s unified GPT-5 line — a general model with configurable reasoning — and is best understood through its coding variant. GPT-5.3-Codex is the agentic coding model designed for long-running software tasks: it reads and writes files, runs commands, works with GitHub and handles multi-step refactors, and OpenAI positioned it as state-of-the-art on SWE-bench Pro at its 24 February 2026 release, while being about 25% faster and more token-efficient than the previous Codex generation.
The general GPT-5.3 model served as a ChatGPT default (“GPT-5.3 Instant”) during early 2026. Because the line was superseded so quickly — by GPT-5.4 on 5 March and GPT-5.5 on 23 April — OpenAI did not leave behind a large public benchmark trail for the general model, so this page leads with the well-documented Codex variant and marks unconfirmed figures as data not available rather than estimating them.
Model variants
| Variant | API ID | What it is |
|---|---|---|
| GPT-5.3 | gpt-5.3 | The general model; powered ChatGPT’s “GPT-5.3 Instant” default |
| GPT-5.3-Codex | gpt-5.3-codex | Agentic coding model (24 Feb 2026); SOTA on SWE-bench Pro at release; $1.75/$14; 400K context |
Benchmark performance
OpenAI described GPT-5.3-Codex as achieving state-of-the-art results on SWE-bench Pro at its February 2026 release, with strong Terminal-Bench 2.0 and OSWorld-Verified scores. Precise headline numbers for the general GPT-5.3 model are not separately published here, so they are left as data not available rather than estimated. For current, fully-benchmarked OpenAI models see GPT-5.4 and GPT-5.5, and the best AI for coding ranking for cross-model coding standings.
Pricing
GPT-5.3-Codex is priced at $1.75 input / $14 output per million tokens with a 400K-token context window. Standard GPT-5.3 per-token pricing was not separately published here; because the line was superseded within two weeks by GPT-5.4 ($2.50/$15), GPT-5.4 and GPT-5.5 are the better current reference points.
How GPT-5.3 compares
- vs GPT-5.4 — GPT-5.4 superseded GPT-5.3 on 5 March 2026, unifying coding, reasoning and computer use and posting the first OSWorld score past the human baseline. For anything other than legacy Codex workflows, GPT-5.4 (or GPT-5.5) is the better choice.
- vs GPT-5.2 — GPT-5.3 was the incremental successor to GPT-5.2 (December 2025), with the Codex variant pushing agentic coding forward.
- vs Claude — On agentic coding, Claude leads the hardest benchmarks; GPT-5.3-Codex was competitive at its release but has since been overtaken within OpenAI’s own line.
Known limitations
Superseded quickly. GPT-5.3 was a frontline model for only a few weeks before GPT-5.4 and then GPT-5.5 replaced it.
Sparse public benchmarks for the general model. OpenAI documented GPT-5.3-Codex well but left fewer headline figures for the general GPT-5.3 model, so several numbers are marked data not available here.
Use the current models instead. For new work, GPT-5.4 (value) or GPT-5.5 (flagship) are the sensible choices; GPT-5.3-Codex remains available mainly for continuity.
Version history
| Version | Released | Key points |
|---|---|---|
| GPT-5.3-Codex | 24 Feb 2026 | Agentic coding; SOTA SWE-bench Pro at release; $1.75/$14; 400K context |
| GPT-5.4 | 5 Mar 2026 | Successor; unified model; 75% OSWorld |
| GPT-5.2 | 11 Dec 2025 | Predecessor |
FAQ
What is GPT-5.3?
GPT-5.3 is an OpenAI model generation from early 2026, best known for GPT-5.3-Codex, its agentic coding variant. The general GPT-5.3 model powered a ChatGPT default (“GPT-5.3 Instant”) before GPT-5.5 Instant replaced it on 5 May 2026.
What is GPT-5.3-Codex?
OpenAI’s agentic coding model, released 24 February 2026. It was described as state-of-the-art on SWE-bench Pro at launch, with strong Terminal-Bench 2.0 and OSWorld-Verified scores, about 25% faster and more token-efficient than prior Codex models. It is priced at $1.75/$14 per million tokens with a 400K context window.
Is GPT-5.3 still current?
No. It was superseded by GPT-5.4 (March 2026) and then GPT-5.5 (April 2026). GPT-5.3-Codex remains available via the API, but new work should use the current models.
Last verified 18 June 2026. GPT-5.3-Codex details are OpenAI-reported from its February 2026 release; figures for the general GPT-5.3 model that could not be confirmed are marked data not available rather than estimated. Pricing and standing change quickly — confirm against OpenAI’s official pages before relying on them.