OpenAI’s model names (e.g., o3, 4o, 4.1, 5.2) are confusing and change often.
This is my quick reference for picking the right model for any task 🧠⚡️
This is a living document - if you spot mistakes or missing models,
please make a PR!
This page lives in GitHub Markdown via TinaCMS 🦙
OpenAI Chat Model Cheat-Sheet (Updated February 2026)
Legend:
- 🟢 current & recommended
- 🟡 supported but not first-choice
- 🔴 legacy / deprecated
| Status | Model | Release / Change | Strengths / Best Use-Cases | Trade-offs / Notes |
| 🟡 | GPT-5.3-Codex | Feb 2026 | Newest Codex flagship for agentic coding + broader professional work on a computer. | Available in Codex for paid ChatGPT plans; API access is not broadly available yet (no stable public API model ID to pin). |
| 🟡 | GPT-5.3-Codex-Spark | Feb 2026 | Ultra-fast, real-time coding (tight interactive loop; minimal targeted edits by default). | Research preview for Codex users; API access limited to select design partners. |
| 🟢 | GPT-5.2 (gpt-5.2) | Dec 2025 | Current flagship “thinking” model for professional work, long-horizon agents, and hard coding/reasoning. | Higher latency/cost; pick smaller variants when you don’t need max capability. |
| 🟢 | GPT-5.2-Codex (gpt-5.2-codex) | 2025-2026 | Most intelligent coding model; optimized for long-horizon, agentic coding tasks (Codex-style). | Coding-specialized; prefer general GPT models for writing/analysis-heavy work. |
| 🟢 | GPT-5.2 Pro (gpt-5.2-pro) | Dec 2025 | Highest-quality GPT-5.2 option when correctness is worth the wait (deep reasoning, fewer major errors). | Much more expensive; overkill for routine tasks. |
| 🟢 | GPT-5.1 (gpt-5.1) | Nov 2025 | Strong general reasoning model with configurable/adaptive reasoning effort. | Usually superseded by 5.2 for frontier work. |
| 🟢 | GPT-5 (gpt-5) | Aug 2025 | First GPT-5 release; still solid for general reasoning/coding. | Generally behind 5.1/5.2 on capability. |
| 🟢 | GPT-5 mini (gpt-5-mini) | 2025-2026 | Faster, cost-efficient GPT-5 family model for well-defined tasks and high throughput. | Less headroom on complex reasoning. |
| 🟢 | GPT-5 nano (gpt-5-nano) | 2025-2026 | Fastest/cheapest GPT-5 family model for simple tasks (classification, routing, autocomplete). | Limited nuance; brittle on long reasoning chains. |
| 🟢 | GPT-4.1 (gpt-4.1) | 14 Apr 2025 | Smartest non-reasoning model; excellent for coding, instruction following, and long context (1M tokens). | Usually weaker on “think hard” problems vs GPT-5.x or o-series reasoning models. |
| 🟢 | GPT-4.1 mini (gpt-4.1-mini) | 14 Apr 2025 | Smaller, faster 4.1; great quality/price default when you don’t need frontier reasoning. | Less capable than full 4.1. |
| 🟢 | GPT-4.1 nano (gpt-4.1-nano) | 14 Apr 2025 | Ultra-fast/cheap; still supports long context; ideal for extraction, tagging, and lightweight copilots. | Loses nuance and robustness. |
| 🟡 | GPT-5.1-Codex-Max (gpt-5.1-codex-max) | 2025-2026 | Codex variant tuned for longer-running agentic coding tasks. | Older generation vs 5.2-Codex; Responses API only. |
| 🟡 | GPT-5.1-Codex (gpt-5.1-codex) | 2025-2026 | Codex-optimized agentic coding model (good default for Codex flows). | Older generation vs 5.2-Codex; Responses API only. |
| 🟡 | GPT-5.1 Codex mini (gpt-5.1-codex-mini) | 2025-2026 | Cheaper/smaller Codex variant for fast coding iterations. | Less capable than larger Codex models. |
| 🟡 | GPT-5-Codex (gpt-5-codex) | 2025-2026 | Earlier Codex family model for agentic coding. | Superseded by newer Codex variants; Responses API only. |
| 🟡 | OpenAI o3 (o3) | Apr 2025 | Strong reasoning model (math/science/coding) and agentic tool use. | Marked as “succeeded by GPT-5” in the API docs; use when you specifically want o-series behavior. |
| 🟡 | OpenAI o4-mini (o4-mini) | Apr 2025 | Cost-efficient reasoning when you want real “thinking” but need speed/volume. | Marked as “succeeded by GPT-5 mini” in the API docs. |
| 🟡 | GPT-4o / GPT-4o mini (gpt-4o, gpt-4o-mini) | 2024-2025 | Versatile multimodal (text + image) general model family; strong for everyday chat + vision. | Not the newest frontier; typically behind GPT-4.1 and GPT-5.x on pure coding/reasoning. |
| 🟡 | GPT-4 Turbo / GPT-4 (gpt-4-turbo, gpt-4) | 2023-2024 | Older but still usable for compatibility and legacy behaviors. | Prefer newer families unless you need an exact legacy behavior/snapshot. |
| 🔴 | GPT-4.5 Preview (gpt-4.5-preview) | Feb 2025 | Research preview model (not intended as a long-term default). | Marked deprecated in the API docs. |
| 🔴 | codex-mini-latest (codex-mini-latest) | 2024-2025 | Deprecated Codex-CLI-focused model (fine-tuned from o4-mini). | Deprecated; docs recommend starting with gpt-4.1 for direct API use. |
| 🔴 | GPT-3.5 Turbo (gpt-3.5-turbo) | 2023-2024 | Legacy cheap chat model for simple tasks. | Outclassed by newer mini/nano models. |
Decoding the Names
- GPT = Generative Pre-trained Transformer
- Major number (3, 4, 5) → generation
- .1 / .2 / .5 → iterative upgrade within a generation
- o-series (e.g.
o3,o4-mini) → reasoning-first models (trained to “think longer”) - 4o → “omni” GPT model family (multimodal inputs)
- mini / nano → smaller, faster, cheaper variants
- pro → more compute for higher-quality answers
- codex → coding-optimized variant (agentic coding focus)
- -chat-latest → ChatGPT-flavored alias (convenience/compat; not always recommended for stable production)
- YYYY-MM-DD in ID (e.g.
gpt-4o-2024-11-20) → a snapshot you can pin for stability
ChatGPT Names vs API Model IDs
ChatGPT’s picker labels don’t always match what you pass to the API.
| ChatGPT label | API model ID |
| ChatGPT-5.2 Instant | gpt-5.2-chat-latest |
| ChatGPT-5.2 Thinking | gpt-5.2 |
| ChatGPT-5.2 Pro | gpt-5.2-pro |