OpenAI Chat Model Cheat Sheet - Decoding the Chaos

OpenAI’s model names (e.g., o4, 4o) are confusing and change often. This guide is my quick reference for picking the best model for any task.

This is a living document and I've probably made some mistakes while writing it, so feel free to suggest changes or additions. I'll keep it updated as OpenAI rolls out new models and features 🚀 If you see something missing, or a mistake, please make a PR! this page is stored as markdown in GitHub using TinaCMS 🦙

OpenAI Chat Model Cheat-Sheet

Status Legend:

🟢 current & recommended
🟡 active but being phased out
🔴 legacy/deprecated

Status	Model	Release date	Strengths / Best use-cases	Trade-offs	Naming note
🟢	GPT-4.1	14 April 2025	Flagship long-context (1M tokens), top-tier coding & RAG	Costliest text-only model	“4.1” = incremental rev over 4
🟢	GPT-4.1 mini/nano	14 April 2025	~80–90% of 4.1 accuracy at lower price/latency	Slightly less reasoning depth	“mini / nano” denote distillations
🟢	GPT-4o	13 May 2024	Native text + image + audio (“omni”) real-time chat	Pricier than 4o mini	“o” = omni multimodal
🟢	GPT-4o mini	July 2024	Cheaper, faster multimodal	Lower accuracy vs 4o	Same naming rule
🟢	o4 mini / o4 mini high	16 April 2025	Efficient reasoning, image-aware; “high” spends more tokens for reliability	Smaller context than 4.1	“mini” = size; “high” = higher reasoning effort setting
🟢	o3	16 April 2025	Deep step-by-step reasoning, code, math, tool use	Slower & pricier than o4 mini	“o-series” = optimized reasoning line
🟢	o3 mini / o3 mini high	February–March 2025	Very cheap STEM reasoning; “high” = extra depth	Outclassed by o4 mini on most tasks	Legacy small variant
🟢	GPT-3.5 Turbo	30 November 2022	Budget workhorse for text	Reasoning weaker than 4-line	“Turbo” = cost/latency optimized snapshot
🟡	GPT-4 Turbo	November 2023	128K context text/vision	Sunset mid-2025	“Turbo” as above
🟡	GPT-3.5 (base)	March–November 2022	Early RLHF model; ChatGPT v1	Small context, dated knowledge	“3.5” signaled RLHF-tuned step between 3 & 4
🔴	GPT-4	14 March 2023	High-quality text; vision via separate endpoint	Retired from ChatGPT 30 April 2025	Plain version number
🔴	GPT-4.5 preview	February 2025	Bridge model while 4o rolled out	API removal slated 14 July 2025	“.5” = half-step; “preview” = experimental
🔴	GPT-3	June 2020	Historic foundation LLM	Lags on reasoning, cut-off 2020	Generation number only
🔴	GPT-2	February 2019	Historic: text generation demo	Small context, safety gaps	Generation number only
🔴	GPT-1	June 2018	Proof-of-concept	117M params, research only	First use of “GPT”

Decoding the names

GPT= Generative Pre-trained Transformer
Major number (1, 2, 3, 4, 4.1…) → new architecture/train run.
.5/.1 → mid-cycle fine-tune refresh.
o → omni (multimodal I/O).
o-series without “GPT” (o1, o3, o4…) → separate reasoning line; letters are branding, numbers track generations.
mini / nano / high → trade speed/cost vs accuracy.
Turbo → snapshot tuned for throughput/cost.
Date format - MMDD (e.g., gpt-4-0613) → training-snapshot date.