GPT-5.6 Luna vs GPT-5.3 Codex
Two OpenAI models with different jobs — a fast, low-cost general tier versus an agentic coding specialist.
Updated August 28, 2026 · Independent comparison — GPT-5.6 Luna, GPT-5.3 Codex are separate products.
GPT-5.6 Luna
OpenAI's cheapest, fastest GPT-5.6 tier
GPT-5.6 Luna is the fastest, most affordable tier of OpenAI's GPT-5.6 family, sitting below the Sol flagship and the Terra mid-tier. It's a reasoning model with a ~1.05M-token context and a 128K output ceiling, and after an 80% price cut on 2026-07-30 it costs just $0.20/$1.20 per million input/output tokens ($0.02 cached input; $0.40/$1.80 in long-context mode). OpenAI positions it for cost-sensitive, high-volume work such as classification, summarization, routing, and real-time apps, and it became the default chat model on the Free and Go plans in early August 2026. It accepts text and image input, supports reasoning effort from none through max, and carries a February 2026 knowledge cutoff.
GPT-5.3 Codex
OpenAI's agentic coding-specialized model
GPT-5.3 Codex is OpenAI's agentic coding specialist, tuned for long-horizon software tasks inside the Codex CLI, the Codex IDE extension, and Codex Cloud. It's priced at $1.75/$14.00 per million tokens with a 90% cache discount ($0.175 cached input); OpenAI has not published its context window or max output. It supports low/medium/high/xhigh reasoning effort, and on Artificial Analysis it posts an intelligence index around 46 and streams at roughly 123 tokens per second, fast for a reasoning coder. It accepts text, images, and files, and carries an August 2025 knowledge cutoff.
Bottom line
Two OpenAI models with different jobs — a fast, low-cost general tier versus an agentic coding specialist. Choose GPT-5.6 Luna if you need the cheapest GPT-5.6 tier for high-volume classification, summarization, routing, or real-time features; choose GPT-5.3 Codex if you're doing agentic, long-horizon coding in the Codex CLI, the IDE extension, or Codex Cloud.
Context & Output
| Feature | GPT-5.6 Luna | GPT-5.3 Codex |
|---|---|---|
| Context window | ~1,050,000 tokens | Not published |
| Max output tokens | 128,000 | Not published |
| Input modalities | Text, image | Text, image, files (PDF) |
| Knowledge cutoff | February 16, 2026 | August 31, 2025 |
Pricing (per 1M tokens)
| Feature | GPT-5.6 Luna | GPT-5.3 Codex |
|---|---|---|
| Input | $0.20 | $1.75 |
| Output | $1.20 | $14.00 |
| Cached input | $0.02 | $0.175 (90% off) |
| Long-context rate | $0.40 / $1.80 | Not published |
Coding & Reasoning
| Feature | GPT-5.6 Luna | GPT-5.3 Codex |
|---|---|---|
| Reasoning model | ✓ | ✓ |
| Reasoning effort levels | none-max (6 levels) | low / medium / high / xhigh |
| Coding-optimized | ✗ | ✓ |
| Artificial Analysis intelligence index | Not published | ~46 |
| Focus | General-purpose / high-volume | Codex agentic coding |
Speed & Latency
| Feature | GPT-5.6 Luna | GPT-5.3 Codex |
|---|---|---|
| Speed tier | Cheapest / fastest GPT-5.6 tier | Fast for a reasoning coder |
| Output speed | Not published | ~123 tokens/sec (Artificial Analysis) |
| Best for | High-volume, low-latency tasks | Long-horizon coding tasks |
Availability & Access
| Feature | GPT-5.6 Luna | GPT-5.3 Codex |
|---|---|---|
| Open weights | ✗ | ✗ |
| OpenAI API | ✓ | ✓ |
| Bundled in | ChatGPT Free/Go default, OpenCode Go | Codex CLI, Codex IDE, Codex Cloud |
| Other surfaces | Azure AI, Amazon Bedrock, Cursor | OpenCode Zen, Cursor |
The Verdict
Choose GPT-5.6 Luna if...
- ✓You need the cheapest GPT-5.6 tier for high-volume classification, summarization, routing, or real-time features.
- ✓You want a large ~1M-token context at $0.20/$1.20 per million tokens.
- ✓Latency and cost per token matter more to you than top-end coding accuracy.
- ✓You want one fast model with adjustable reasoning effort from none to max.
- ✓A general-purpose model is fine — you don't need a coding specialist.
Choose GPT-5.3 Codex if...
- ✓You're doing agentic, long-horizon coding in the Codex CLI, the IDE extension, or Codex Cloud.
- ✓You want a model tuned for tool use and multi-step software tasks, not just chat.
- ✓You value the 90% cache discount on repeated large-context coding sessions.
- ✓You want fast token streaming (~123 t/s) with high reasoning effort available.
- ✓Coding accuracy justifies the higher $1.75/$14 per-million pricing.
Frequently Asked Questions
What is the difference between GPT-5.6 Luna and GPT-5.3 Codex?
GPT-5.6 Luna is the fastest, most affordable tier of OpenAI's GPT-5.6 family, sitting below the Sol flagship and the Terra mid-tier. GPT-5.3 Codex is OpenAI's agentic coding specialist, tuned for long-horizon software tasks inside the Codex CLI, the Codex IDE extension, and Codex Cloud.
How much do GPT-5.6 Luna and GPT-5.3 Codex cost?
GPT-5.6 Luna: $0.20. GPT-5.3 Codex: $1.75. See the pricing table above for full plan details.
Should I choose GPT-5.6 Luna or GPT-5.3 Codex?
Choose GPT-5.6 Luna if you need the cheapest GPT-5.6 tier for high-volume classification, summarization, routing, or real-time features. Choose GPT-5.3 Codex if you're doing agentic, long-horizon coding in the Codex CLI, the IDE extension, or Codex Cloud.
1DevToolRun OpenAI's models on your terms
Both GPT-5.6 Luna and GPT-5.3 Codex ship through OpenAI's API and Codex tooling. 1DevTool lets you drive them from the Codex CLI — alongside Claude Code, Gemini CLI, Cline, and OpenCode — in persistent terminals: bring your own OpenAI key or ChatGPT plan, switch models per task, and use the built-in HTTP client, 13-engine database client, and embedded browser. One-time $29, no subscription.