GPT (OpenAI)
Overview
The GPT series is OpenAI's family of large language models. Since ChatGPT launched in 2022 it has been the de facto standard of the LLM market, reachable through the API, ChatGPT, and a wide range of products. As of August 2026 the flagship is GPT-5.6 (general availability July 9, 2026), reorganized into three tiers: Luna, Terra, and Sol.
Current model: GPT-5.6
GPT-5.6 reached general availability on July 9, 2026. The previous five-variant lineup was consolidated into Luna / Terra / Sol (least to most capable), making the mapping between price and use case much easier to reason about.
The three tiers and API pricing
On July 30, 2026 — three weeks after GA — OpenAI cut prices on the two lower tiers: Luna input dropped from $1.00 to $0.20 (an 80% cut) and Terra input from $2.50 to $2.00 (20%). Sol is unchanged since launch.
| Model | Best for | API pricing (input / output per 1M tokens) |
|---|---|---|
| GPT-5.6 Sol | Top capability: hard reasoning, research, agents | $5 / $30 |
| GPT-5.6 Terra | General-purpose flagship | $2 / $12 |
| GPT-5.6 Luna | Fast and cheap, high-volume workloads | $0.20 / $1.20 |
What shipped in August 2026
- August 10 — GPT-5.6-Cyber: a cybersecurity-focused model built on Sol, offered through a trusted-access program for critical-infrastructure defense and vulnerability research.
- August 13 — Ultrafast mode (preview): runs GPT-5.6 Sol at up to 14x the usual speed. Useful for interactive work and for agents where you want more attempts per minute.
Key capabilities
1. Adjustable reasoning effort
Answer simple questions immediately, or spend more time thinking through hard ones. Controlling this yourself is how you balance cost against quality.
2. Native Computer Use API
An autonomous agent capability that reads the screen and performs clicks and keystrokes, exposed directly through the API. It automates browser work, desktop apps, and form filling, and absorbed the earlier Operator product.
3. 1M+ token context window
Handles long documents, whole codebases, and multi-file references.
4. Codex (coding agent)
Codex, once a separate coding-specialized model, is folded into the main line, so frontier-level code generation is available for general use too. Codex also runs in the ChatGPT apps for iOS and Android, and a partnership with Dell brings it to hybrid and on-premises environments.
5. GPT Image 2.0 (image generation)
Released April 21, 2026 and billed by tokens: $5 per 1M text input, $8 per 1M image input, and $30 per 1M image output. Per 1024×1024 image that works out to roughly $0.006 at low quality, $0.053 at medium, and $0.211 at high.
6. Sora (video generation) — discontinued
Sora has been shut down. OpenAI announced the discontinuation on March 24, 2026; the web, iOS and Android apps closed on April 26, 2026, and the Sora 2 API is removed on September 24, 2026. No successor has been announced. If you are on the API, you need a migration target before September 24.
The stated reason is economics: Sora reportedly cost around $1M per day to run against roughly $2.1M in total revenue. OpenAI is redirecting compute toward coding and enterprise products. Runway, Kling and Luma are the usual migration targets for video generation.
ChatGPT plans
| Plan | Monthly | What you get |
|---|---|---|
| Free | $0 | Limited access to GPT-5.6 (ad-supported) |
| Go | $8 | More headroom than Free; ad-supported |
| Plus | $20 | The GPT-5.6 family, image generation, voice mode, Deep Research (10 runs/mo), Codex, Agent Mode |
| Pro | $100 | Roughly 5x the usage of Plus. Includes Codex (launched April 9, 2026) |
| Pro | $200 | Roughly 20x the usage of Plus, the Operator agent, priority access |
| Business | $25/user | Team features, admin console, data protection |
| Enterprise | Contact sales | SSO, audit logs, dedicated support, SLA |
Pro comes in $100 and $200 flavors. The difference is headroom (roughly 5x versus 20x Plus), so $100 is enough unless you are running agents continuously.
Typical uses
- Software development: Codex, Canvas, code execution, multi-file edits, debugging
- Research: Deep Research, long-document summarization, paper analysis
- Workflow automation: autonomous PC operation via Computer Use
- Content generation: articles, email, marketing copy
- Data analysis: Advanced Data Analysis, Python execution, charting
- Image generation: GPT Image 2.0 (Sora, the video model, has been shut down)
Strengths
- ✅ Unified architecture — reasoning, code, agents, and multimodality in one model
- ✅ Computer Use API — native autonomous control of a computer
- ✅ Largest ecosystem — GPTs marketplace, plugins, extensive SDKs
- ✅ Industry-standard API — REST plus Python/Node/Java SDKs
- ✅ A real price ladder — at $0.20 per 1M input tokens, Luna is cheap enough for bulk workloads
- ✅ Multimodal — text, image, audio, video, Computer Use
Weaknesses
- ❌ The upper Pro tier costs $200/month
- ❌ Sol's API pricing ($5/$30) is 2.5x Terra's, so routing everything to it gets expensive fast
- ❌ Closed source (weights are not published)
- ❌ API rate limits at peak times
- ❌ Privacy considerations for enterprise data
Official links
- Website: https://openai.com
- API docs: https://platform.openai.com/docs
- ChatGPT: https://chatgpt.com
- Pricing: https://openai.com/api/pricing