GPT-5.3-Codex in 2026: context, pricing, reasoning, agentic coding, and current status
Updated: Sep 17
GPT-5.3-Codex remains a current specialized coding model in 2026, with a 400,000-token context window, 128,000-token maximum output, and $1.75 input / $14 output per million tokens.
It was launched on February 5, 2026 as a model that combines the GPT and Codex training stacks for long-running software-development work, research, tool use, and complex execution.
··········
GPT-5.3-CODEX IS A SPECIALIZED AGENTIC CODING MODEL
OpenAI positions GPT-5.3-Codex around agentic coding rather than ordinary chat. It is designed to plan, edit code, use tools, research, and continue across longer technical workflows.
........
Specification | GPT-5.3-Codex |
|---|---|
Context window | 400,000 tokens |
Maximum output | 128,000 tokens |
Knowledge cutoff | August 31, 2025 |
Reasoning effort | low, medium, high, xhigh |
Input price | $1.75 per 1M tokens |
Cached input | $0.175 per 1M tokens |
Output price | $14 per 1M tokens |
........
··········
THE MODEL COMBINES CODING AND GENERAL-PURPOSE REASONING
OpenAI describes GPT-5.3-Codex as the first model combining its Codex and GPT-5 training stacks. The design goal is broader professional computer work rather than narrow autocomplete or single-file code generation.
··········
400K CONTEXT IS LARGE, BUT SMALLER THAN GPT-5.6 AND ASTRA
GPT-5.3-Codex lists 400K context versus 1.05M for GPT-5.6 Sol, Terra, Luna and GPT-6 Astra. That does not make GPT-5.3-Codex obsolete: context size is only one design dimension, while its agentic coding behavior is specialized.
··········
REASONING CAN SCALE FROM LOW TO XHIGH
The model supports low, medium, high and xhigh reasoning effort. Increasing effort can help difficult planning and debugging but generally increases latency and token use, so the optimal setting depends on task complexity.
··········
DS CALCULATION: A 10M INPUT + 2M OUTPUT WORKLOAD
At list price, 10 million uncached input tokens cost $17.50 and 2 million output tokens cost $28, for a total of $45.50 before any tool-specific charges. Cached input can reduce the input portion materially when reuse is high.
··········
GPT-5.3-CODEX NOW COEXISTS WITH GPT-5.6 AND ASTRA IN CODEX
Codex's product model menu has expanded beyond dedicated Codex-branded models. GPT-5.6 Terra, Sol and Luna and GPT-6 Astra now cover other cost and capability points, while GPT-5.3-Codex remains a specialized option.
··········
WHEN GPT-5.3-CODEX STILL MAKES SENSE
It remains relevant when a workflow benefits from a model explicitly optimized for agentic software engineering, especially when compatibility, established prompts, cost profile, or behavior has already been validated against this model.
··········
FOLLOW US FOR MORE
··········
DATA STUDIOS
··········
