OpenAI Models Landscape — Complete Reference & Cost Estimator
A full breakdown of every model in OpenAI’s current API portfolio — organized by tier, use case, and cost — with an interactive calculator to estimate your production spend before you build.
Four Layers. One Portfolio.
OpenAI folded standalone reasoning models into the flagship line in 2026. Understanding which layer — and which reasoning effort setting — fits your workload now matters more than knowing every model name.
Frontier General
GPT-5.6 Sol — highest capability, adjustable reasoning effort (none → max), 1M+ token context, premium pricing.
Value Production
GPT-5.6 Terra, GPT-5.6 Luna — high-volume defaults, fast, cost-efficient, same 1.05M context window as Sol.
Reasoning Effort (built-in)
o3 and o4-mini were retired Feb 2026. Set reasoning_effort on any GPT-5.6 model instead of switching models.
Specialist Media
GPT Image 2, Sora 2, gpt-realtime-2.1, gpt-4o-transcribe — modality-native models for image, video, voice, transcription.
Compatibility
GPT-5.5, GPT-5.4 family — still fully priced and supported, useful for cost comparison, not recommended for new builds.
Cadence Has Stayed Aggressive
OpenAI has shipped a new flagship family roughly every one to two months since GPT-5’s August 2025 launch.
What Changed Since Cohort 10 (May 2026)
Reasoning Effort Replaces Standalone Reasoning Models
o3 and o4-mini are retired. Every GPT-5.6 model now accepts a reasoning_effort parameter (none, low, medium, high, xhigh, max) — one model line spans everyday chat and deep, multi-step reasoning.
1.05M Token Context Is Now the Default
GPT-5.6 Sol, Terra, and Luna all ship with a 1.05M token context window out of the box — no separate “long context” model needed.
Four Pricing Tiers Per Model
Standard, Batch, Flex, and the newly renamed Fast mode (formerly Priority processing, renamed 30 Jul 2026) each carry different multipliers — batch/flex run roughly half of standard, fast mode roughly double.
Regional Data Residency Carries a 10% Uplift
Models released on or after 5 Mar 2026 that support regional processing endpoints are billed a 10% premium over standard pricing for data residency.
Modality-Native Models Still Win Their Domains
Voice (gpt-realtime-2.1), image (GPT Image 2), video (Sora 2), and transcription (gpt-4o-transcribe, gpt-transcribe) remain dedicated specialist models — don’t route audio or image work through general GPT endpoints.
The Models Worth Knowing
Not the full 70+ catalog — the models that actually show up in production builds.
Frontier model for complex reasoning, coding, and agentic tool use. 1.05M context, reasoning effort none→max.
Balances intelligence and cost. Same 1.05M context and tool support as Sol, roughly 60% cheaper.
Cost-sensitive, high-throughput workloads — classification, tagging, lightweight chat at scale.
Still fully supported, priced comparably to GPT-5.6. No longer the default recommendation for new builds.
State-of-the-art image generation and editing. Text and image inputs, image output billed per token.
Text-to-video generation, billed per second of output. Pro adds 1024p and 1080p resolutions.
Flagship voice model for live, low-latency audio-in/audio-out agents with tool use.
High-accuracy speech-to-text for file and realtime transcription.
Best cost/quality embedding model for retrieval, RAG, and search. 1,536 dimensions.
Text & Reasoning — Standard Pricing
Prices per 1M tokens, Standard tier, short context. Long-context requests run roughly 2× these rates on GPT-5.6/5.5/5.4.
| Model | Input | Cached input | Output | Status |
|---|---|---|---|---|
| gpt-5.6-sol | $5.00 | $0.50 | $30.00 | Flagship — recommended |
| gpt-5.6-terra | $2.00 | $0.20 | $12.00 | Balanced — default |
| gpt-5.6-luna | $0.20 | $0.02 | $1.20 | High-volume |
| gpt-5.5 | $5.00 | $0.50 | $30.00 | Legacy — still supported |
| gpt-5.5-pro | $30.00 | — | $180.00 | Legacy pro tier |
| gpt-5.4 | $2.50 | $0.25 | $15.00 | Legacy |
| gpt-5.4-mini | $0.75 | $0.075 | $4.50 | Legacy |
| gpt-5.4-nano | $0.20 | $0.02 | $1.25 | Legacy |
| gpt-5.4-pro | $30.00 | — | $180.00 | Legacy pro tier |
Batch tier ≈ 50% of Standard. Fast mode ≈ 2× Standard. Regional data-residency endpoints add a 10% uplift on eligible models (released ≥ 5 Mar 2026).
| Model | Modality | Input | Cached input | Output |
|---|---|---|---|---|
| gpt-realtime-2.1 | Audio | $32.00 | $0.40 | $64.00 |
| ″ | Text | $4.00 | $0.40 | $24.00 |
| ″ | Image | $5.00 | $0.50 | — |
| gpt-realtime-2.1-mini | Audio | $10.00 | $0.30 | $20.00 |
| ″ | Text | $0.60 | $0.06 | $2.40 |
| gpt-realtime-translate | Audio | — | $0.034/min | |
| gpt-live-transcribe | Audio | — | $0.017/min | |
| Model | Modality | Input | Cached input | Output |
|---|---|---|---|---|
| gpt-image-2 | Image | $8.00 | $2.00 | $30.00 |
| ″ | Text | $5.00 | $1.25 | — |
| gpt-image-1.5 | Image | $8.00 | $2.00 | $32.00 |
| gpt-image-1-mini | Image | $2.50 | $0.25 | $8.00 |
| Model | Size | Price / second |
|---|---|---|
| sora-2 | 720p | $0.10 |
| sora-2-pro | 720p | $0.30 |
| ″ | 1024p | $0.50 |
| ″ | 1080p | $0.70 |
| Model | Estimated cost |
|---|---|
| gpt-transcribe | $0.0045 / minute |
| gpt-live-transcribe | $0.017 / minute |
| gpt-4o-transcribe | $0.006 / minute ($2.50 in / $10.00 out per 1M tokens) |
| gpt-4o-mini-transcribe | $0.003 / minute ($1.25 in / $5.00 out per 1M tokens) |
| Model | Price / 1M tokens | Dimensions | Best for |
|---|---|---|---|
| text-embedding-3-small | $0.02 | 1,536 | Retrieval, RAG, search — best cost/quality |
| text-embedding-3-large | $0.13 | 3,072 | Max precision use cases |
| Tool | Details | Pricing |
|---|---|---|
| Web search | All models | $10.00 / 1k calls + content tokens at model rates |
| ″ | Non-reasoning preview | $25.00 / 1k calls, content tokens free |
| Containers | Hosted Shell & Code Interpreter | $0.03–$1.92 per 20-min session (1GB–64GB) |
| File search | Storage | $0.10/GB/day (1GB free) |
| ″ | Tool call | $2.50 / 1k calls |
Estimate Your Production Spend
Pick a model and enter your expected monthly token volume to estimate cost before you build.
Estimate only — excludes cached-input discounts, tool-call fees, and long-context multipliers. Always confirm against OpenAI’s live pricing page before committing budget.
What Should I Use For…
The fastest way to pick a model: start from the task, not the model name.
How To Choose in Five Questions
Is this a build/prototype or a production workload?
Prototyping → default to GPT-5.6 Terra and iterate. Production at scale → model your monthly token volume with the calculator above before committing.
Does the task need deep, multi-step reasoning?
If yes, use GPT-5.6 Sol with reasoning_effort set to high or xhigh. If the task is straightforward (chat, drafting, simple Q&A), leave effort at none or low — you’re paying for reasoning tokens either way.
Is the workload high-volume and cost-sensitive?
Classification, tagging, and simple summarization at scale → GPT-5.6 Luna. For non-urgent batch jobs, stack the Batch tier discount (≈50% off) on top.
Is the input non-text (voice, image, video)?
Route to the modality-native specialist model — gpt-realtime-2.1 for voice, GPT Image 2 for images, Sora 2 for video, gpt-4o-transcribe for transcription — never through a general chat model.
Does latency matter more than cost right now?
Use Fast mode (formerly Priority processing, renamed 30 Jul 2026) — roughly 2× Standard pricing for faster response times. Otherwise, Standard tier is the right default.
Decoding Data Science
Prepared for Cohort 11 · AI Residency · Pricing as of 1 August 2026 · Source: OpenAI official docs & pricing page