WritingmateWritingmate

AI model reference · 2026

GPT-5.3 Codex

OpenAI: GPT-5.3-Codex logoOpenAIModel details, evidence, and API facts

GPT-5.3-Codex is OpenAI’s most advanced agentic coding model, combining the frontier software engineering performance of GPT-5.2-Codex with the broader reasoning and professional knowledge capabilities of GPT-5.2.

Text generation
Image input
Reasoning
Provider endpoint accepts tool parameters
Try GPT-5.3 CodexCompare with another model
Context window
400,000 tokens
Maximum output
128,000 tokens
Base provider API input
$1.75 / 1M tokens
Base provider API rate
Published evidence
8 benchmarks, 0 Arena results

Overview

About GPT-5.3 Codex

A concise catalog overview. Technical limits and published evaluation evidence are listed separately below.

GPT-5.3-Codex is OpenAI’s most advanced agentic coding model, combining the frontier software engineering performance of GPT-5.2-Codex with the broader reasoning and professional knowledge capabilities of GPT-5.2.

text
multimodal
reasoning
coding

Published evaluations

Benchmark results

Only results mapped to this exact model and backed by a named source appear here. Different evaluation protocols are not treated as interchangeable.

Source-backed benchmark results for GPT-5.3 Codex
BenchmarkScoreEvaluation detailsSource
BrowseCompA benchmark of difficult, verifiable information-seeking questions designed to measure an agent's ability to locate hard-to-find facts through web browsing.77.3%AccuracyVersion: BrowseComp · Tools: Web browsing · Reasoning effort: xhighMethodology OpenAIProvider reported
OSWorld-VerifiedA verified computer-use benchmark in which multimodal agents operate desktop applications and are graded from the resulting environment state.74.0%Mean task rewardVersion: OSWorld-Verified · Reasoning effort: xhighProtocol note: Updated result using the API parameter that preserves original image resolution; supersedes OpenAI's 64.7 launch resultMethodology OpenAIProvider reported
SWE-Bench ProA contamination-resistant software-engineering benchmark with long-horizon tasks across multiple programming languages. Public and private splits are distinct protocols.56.8%ResolvedVersion: SWE-Bench Pro · Split: 731 public tasks · Reasoning effort: xhighMethodology OpenAIProvider reported
SWE-Lancer (IC-Diamond subset)The individual-contributor Diamond subset of SWE-Lancer, which evaluates economically valuable real-world software-engineering tasks.81.4%Tasks completedSplit: IC Diamond · Reasoning effort: xhighMethodology OpenAIProvider reported
Terminal-Bench 2.0Version 2.0 of the benchmark for completing realistic tasks in terminal environments. Results must not be merged with Terminal-Bench 2.1.77.3%Pass rateVersion: 2.0 · Reasoning effort: xhighMethodology OpenAIProvider reported
Terminal-Bench 2.1Version 2.1 of the benchmark for completing realistic tasks in terminal environments. Harness, resource limits, and attempt count are part of the protocol.79.1%Mean task successVersion: 2.1 · Harness: Codex CLIProtocol note: Verified owner-leaderboard run; not protocol-compatible with Meta's provider-reported bash-tool-only harnessMethodology Terminal-BenchBenchmark owner
GPQA DiamondThe highest-quality subset of Graduate-Level Google-Proof Q&A, designed to test expert-level scientific reasoning in biology, physics, and chemistry.92.6%AccuracyVersion: GPQA Diamond · Reasoning effort: xhighMethodology OpenAIProvider reported
Cybersecurity CTFsCapture-the-flag challenges used to evaluate an agent's practical cybersecurity task performance.77.6%Challenges completedReasoning effort: xhighOpenAIProvider reported

Human preference

Arena results

Arena scores come from blind human preference votes. They are reported separately from task benchmarks and are not used as substitutes for missing benchmark results.

Published Arena rows for webdev are conflicting or ambiguous; no score was selected.

Technical reference

Specifications and API facts

Technical limits and reference provider prices come from the current catalog unless a reviewed external source is listed. Missing values stay marked as unavailable.

Model
Provider
OpenAI
Release date
Feb 5, 2026
Catalog added
Feb 24, 2026
Context window
400,000 tokens
Maximum output
128,000 tokens
Knowledge cutoff
Not publicly disclosed
License
Proprietary
Capabilities and API
Input types
Text, Image, File
Output types
Text
Reasoning
Supported
Provider endpoint tool parameters
Accepts tool parameters
Base provider API input
$1.75 / 1M tokens
Base provider cached input
$0.175 / 1M tokens
Base provider API output
$14 / 1M tokens

These are reference provider API rates, not Writingmate checkout charges. Access in Writingmate follows the allowances of your Writingmate plan.

Provenance

Sources and update status

Source links are attached to the facts and evaluations they support. Catalog-only values are not presented as independently verified claims.

Evidence last updated Aug 25, 2026.

Catalog record updated Feb 24, 2026.

Catalog-added dates describe when a model entered the catalog, not necessarily its public release date.

Try GPT-5.3 Codex in Writingmate

Use this model, then compare its response with other AI models in the same workspace.

Get startedCompare models