OpenAI's current flagship is GPT-6 Astra, announced on September 3, 2026 and broadly available from September 4. Its most recent release is GPT-6.1 Sol, which shipped on September 29, 2026 at DevDay. Between the two, GPT-6 Sol and GPT-6 Luna arrived on September 22. This page lists the main GPT releases from GPT-4 onward, with context windows and knowledge cutoffs taken from OpenAI's model pages and shutdown dates from its deprecations page.
GPT Model Timeline
| Model | Released | Context window | Notable change | Status |
|---|---|---|---|---|
| GPT-4 | March 14, 2023 | 8K, with a 32K API variant | Successor to GPT-3.5 | Replaced in ChatGPT in April 2025. API snapshot shuts down October 23, 2026 |
| GPT-4 Turbo | November 2023 | 128K | Sixteen times the default GPT-4 window | API shutdown October 23, 2026 |
| GPT-4o | May 13, 2024 | 128K | Text and image in one model | Retired from ChatGPT on April 3, 2026. First API snapshot shuts down October 23, 2026 |
| GPT-4.5 | February 27, 2025 | Not disclosed on current pages | A particularly large model | API access ended July 14, 2025 |
| GPT-4.1 | April 14, 2025 | 1M | 1M window. No reasoning step | Removed from ChatGPT in February 2026. Still on the API |
| o3 | April 16, 2025 | 200K | Reasoning model, successor to o1 | Retired from ChatGPT on August 26, 2026. API shutdown December 11, 2026 |
| GPT-5 | August 7, 2025 | 400K | Quick and reasoning modes in one model | API shutdown December 11, 2026 |
| GPT-5.1 | November 2025 | 400K | Same knowledge cutoff as GPT-5 | Chat alias shut down July 23, 2026 |
| GPT-5.2 | December 11, 2025 | 400K | Instant, Thinking, and Pro modes | Discontinued in ChatGPT on June 12, 2026 |
| GPT-5.3 Codex | February 5, 2026 | 400K | Coding model. An Instant variant followed | GPT-5.3 chat alias shut down August 10, 2026 |
| GPT-5.4 | March 5, 2026 | 1,050,000 | First GPT-5 model with a 1M-class window | On the API |
| GPT-5.5 | April 23, 2026 | 1,050,000 | Long-running agent work | On the API |
| GPT-5.6 Sol, Terra, Luna | July 9, 2026 | 1,050,000 | Named tiers replace base, mini, and nano. Sol preview from June 26 | Active. Luna serves free ChatGPT users |
| GPT-6 Astra | September 3, 2026 | 1,050,000 | First GPT-6 model | Current flagship |
| GPT-6 Sol, GPT-6 Luna | September 22, 2026 | 1,050,000 | Mid and low-cost tiers at about half the GPT-5.6 price | Active. Sol superseded by 6.1 |
| GPT-6.1 Sol | September 29, 2026 | 1,050,000 | Near-Astra results at one fifth of the price | Latest release |
Knowledge Cutoffs and API Prices
| Model | Knowledge cutoff | Input / output per million tokens |
|---|---|---|
| GPT-4 Turbo | December 1, 2023 | $10 / $30 |
| GPT-4o | October 1, 2023 | $2.50 / $10 |
| GPT-4.1 | June 1, 2024 | $2 / $8 |
| o3 | June 1, 2024 | $2 / $8 |
| GPT-5 and GPT-5.1 | September 30, 2024 | $1.25 / $10 |
| GPT-5.2 | August 31, 2025 | $1.75 / $14 |
| GPT-5.4 | August 31, 2025 | $2.50 / $15 |
| GPT-5.5 | December 1, 2025 | $5 / $30 |
| GPT-5.6 Sol | February 16, 2026 | $4 / $20 |
| GPT-6 Sol | April 20, 2026 | $2 / $10 |
| GPT-6 Astra | April 30, 2026 | $10 / $50 |
| GPT-6.1 Sol | April 30, 2026 | $2 / $10 |
| GPT-6 Luna | May 18, 2026 | $0.10 / $0.50 |
The cutoff does not always move forward with the version number. GPT-4o's cutoff is earlier than GPT-4 Turbo's, GPT-5.1 kept GPT-5's, and GPT-5.4 kept GPT-5.2's. From GPT-5.5 onward, the gap between cutoff and launch has been four to five months. From GPT-5.4 onward, prompts above 272,000 input tokens are billed at twice the input rate and 1.5 times the output rate.
Which ChatGPT Plan Runs Which Model
Free and Go users have been on GPT-5.6 Luna since August 2026. Requesty reports that GPT-6 Luna reached those plans only in the desktop app at launch. Plus and higher plans get GPT-6 Astra, on a staged rollout, and GPT-6.1 Sol in ChatGPT Work and Codex. Dots, the always-on agents on Pro and Business Premium, run on Astra. Details are on the DevDay 2026 page.
Retirements Still Ahead
Two API shutdown dates remain in 2026. On October 23, the original GPT-4, GPT-4 Turbo, the first GPT-4o snapshot, o1, o3-mini, and o4-mini go dark. On December 11, the GPT-5 launch snapshots and o3 follow. The full calendar is in the AI model deprecation tracker.
What Is Next
GPT-6.1 Astra. It was expected at DevDay and did not ship. Reporting from WIRED and the Wall Street Journal says OpenAI held it back after it missed an internal safety bar. OpenAI has not given a new date.
Ultrafast for GPT-6.1 Sol. OpenAI lists a faster serving option as planned.
OpenAI has not announced any model beyond the GPT-6.1 family.
Brand Visibility Implications
- Each cutoff jump changes unprompted recall. GPT-5.5 to GPT-6 Astra moved the cutoff from December 2025 to April 2026. Coverage a brand earned in that window is now in the model. Coverage after April 2026 still depends on search.
- Free users are on an older model. Most ChatGPT users see GPT-5.6 Luna, with a February 2026 cutoff for the family. Paid users see Astra. The same question can return different brands by plan.
- API products lag further. GPT-4o and GPT-5 snapshots stay callable until late 2026, so third-party assistants may still answer from 2023 or 2024 knowledge.
See the training cutoff tracker and how cutoffs change brand recall.
Methodology
This page is built from vendor documentation and dated reporting, not from Presenc AI measurements. Primary sources are used wherever one exists: OpenAI's model pages and deprecations page, Google's Gemini API changelog and deprecations page, xAI's release notes and model list, DeepSeek's API changelog and pricing page, and MiniMax's pricing page and Hugging Face model cards. Older release dates that vendors no longer document come from the relevant Wikipedia articles. Where a figure comes only from a secondary tracker or a single outlet, the page names it. "Not disclosed" means the vendor material we read does not state the figure. Status as of October 1, 2026.
How Presenc AI Helps
Presenc AI tracks how ChatGPT describes and recommends your brand and records the model behind each answer. When OpenAI changes the default model or retires an old one, your history stays comparable.