GPT-5.6 Leaks and What We Know About OpenAI's Next Model

01Five key points

• OpenAI has not announced GPT-5.6. As of June 18, 2026, the name appears on no official page, with no API ID and no published benchmark.
• The main trace comes from internal Codex logs, where the ID gpt-5.6 briefly appeared before May 13, 2026, then disappeared.
• Prediction markets gave 80 to 89% odds of a public release before June 30, 2026.
• The leaks point to a context window of 1.5 million tokens and better performance on long tasks.
• No price is confirmed. The Wall Street Journal reports price cuts coming to OpenAI's API.

0202Why the GPT-5.6 name is already circulating

Barely seven weeks after GPT-5.5 came out on April 24, 2026, the GPT-5.6 name is already all over developer forums. It all started with a technical detail. The researcher Haider spotted an entry in the infrastructure of Codex, OpenAI's coding tool, that mentioned the ID gpt-5.6. The reference could be reproduced for a few hours, then vanished from later session files, which only showed gpt-5.5. This kind of trace often points to a very limited internal test, the sort of experimental rollout a vendor runs before a release.

Developer logs also revealed code names, ember-alpha and beacon-alpha, with the second one linked to long-context tests. Prediction markets such as Polymarket followed suit. In mid-May they showed an 80 to 89% probability of a public release before June 30. The most cited window runs from June 15 to July 5, 2026. All of this remains a bettors' estimate, while the vendor itself says nothing for now.

0303What the leaks suggest

The leaks converge on three areas, all aimed at professional use rather than consumer chat. The first is context size. Several Pro users report behavior consistent with 1.5 million input tokens, 43% more than the 1 million GPT-5.5 accepts. OpenAI has not documented this value. It comes from observed usage, since no technical sheet has been published.

The second area is autonomous work. The leaks describe longer tool chains with less supervision and reliability aimed at agentic sessions lasting several hours, in line with what the Claude Fable 5 model already offers. The third would be a very fast completion mode in code editors, built to compete with the assistants developers use every day. OpenAI is also said to be aiming for a 10 to 15% efficiency gain in the number of tokens used for the same task.

What the leaks attribute to GPT-5.6 compared with models already available (figures still to be confirmed).

Model	Context window	Status
GPT-5.6 (leaks)	~1.5M tokens	unconfirmed
GPT-5.5	1M tokens	available
Claude Fable 5	1M tokens	suspended since June 12
Gemini 3.5 Pro	~2M tokens	announced

This table helps place GPT-5.6 in the current race, but it calls for a warning. Window size says nothing about the model's real quality over such a large context. A model can accept two million tokens and still lose the thread halfway through a long document. Only independent tests, once the model is released, will show what these numbers are worth in practice.

0404The pricing question

This is the point that matters most if you pay per use. Today, the standard version of GPT-5.5 costs $5 per million input tokens and $30 per million output tokens. The Wall Street Journal reports that OpenAI is preparing aggressive price cuts on its API. According to several leaks, GPT-5.6 could be priced below Claude Fable 5, which costs $10 for input and $50 for output, to win back the lead in autonomous agents.

If these cuts are confirmed, the effect will reach well beyond GPT-5.6. Every time a major model gets cheaper, the others follow, as recent consumer subscription cuts showed. If you automate tasks, your final bill depends as much on the price per token as on the volume you use, as GitHub Copilot's token billing made clear. Cutting that volume remains the most reliable lever, whatever model you pick.

Update of October 16, 2026. OpenAI released the GPT-5.6 family on July 9, 2026, as three models named Sol, Terra and Luna. Its official price list now shows Sol at $4 per million input tokens and $20 per million output tokens (a promotional rate announced through at least November 21, 2026), Terra at $2 and $12 and Luna at $0.20 and $1.20. Sol is therefore priced below Claude Fable 5, which has been available again since July 1. GPT-5.5 keeps its price of $5 and $30.

0505What could delay the release

Bettors' odds are only odds. Two obstacles could push back the release of GPT-5.6. One is the quiet period required by the SEC. OpenAI has filed for an IPO, which sharply limits its marketing communication for several weeks. Launching a flagship model in the middle of a regulatory quiet period is tricky. The company may prefer to wait.

Computing power is the other obstacle. Every new flagship model needs a lot of servers. OpenAI has already had to ration access to its recent models at launch. A release could easily slip from June to July or even to late summer without signaling any deeper problem. Caution makes sense as long as there is no official page.

0606What it changes if you work with AI

You don't need to wait for GPT-5.6 to start a project. The tools available today already cover most of what you need to build or optimize websites. A new flagship model mainly changes two things. It shifts the balance between cost and quality on heavy tasks. It also puts downward pressure on the prices of every other model on the market.

The right habit does not depend on the model of the moment, because it is about picking the right tool for each task. A meta description or a short answer does not need the most expensive model. A site audit covering hundreds of pages, on the other hand, justifies a model with stamina. Testing a model on release day doesn't even require a subscription, since you only need to get an API key from the provider.

One last habit applies to any new model. It concerns data control. When you use your own key, your key stays with you and calls go straight to the provider, with no middleman storing your requests. Also set a spending cap before you launch long tasks, because output tokens go fast on a model of this class.

0707How to follow the story without getting fooled

The topic is hot, so misleading headlines will multiply in the coming days. Three simple markers help you separate news from rumor. First, the only source that counts for a new model is OpenAI's official page along with its API documentation. Until the ID appears there, the model has not really shipped, whatever the screenshots say.

Next, be wary of benchmarks published without a link to a reproducible method, because a score quoted out of context is worth very little. Finally, remember that code names often change before release, so the final model may not be called GPT-5.6 at all. Nothing is guaranteed until OpenAI publishes its own announcement, with an ID and prices in black and white.

0808Frequently asked questions

Is GPT-5.6 already available?

No. As of June 18, 2026, OpenAI has published no announcement, no API ID and no benchmark for GPT-5.6. The information going around comes from developer leaks and prediction markets, while the company itself has said nothing.

When will GPT-5.6 be released?

There is no official date so far. Prediction markets expected a release before the end of June 2026, with a wider window into early July. OpenAI's IPO filing could still push this timeline back by several weeks.

What would GPT-5.6 add over GPT-5.5?

The leaks mention a context window raised to 1.5 million tokens, better performance on long autonomous tasks and a fast completion mode for code. All of this still needs to be checked on the day the model actually ships.

Will GPT-5.6 be cheaper?

The Wall Street Journal reports price cuts coming to OpenAI's API. Some leaks place GPT-5.6 below the price of Claude Fable 5. No price is confirmed for now. For reference, GPT-5.5 costs $5 per million input tokens and $30 per million output tokens.

0909References

1. OpenAI, official newsroom, accessed June 18, 2026 (no page about GPT-5.6 on that date).
2. TokenMix, "GPT-5.6 Release Date: Codex Leaks, June Odds, What's Real", 2026.
3. andrew.ooo, "GPT-5.6 Leaked Features: What to Expect From OpenAI's June 2026 Release", 2026.
4. WaveSpeed, "GPT-5.6 Just Showed Up in OpenAI's Codex Logs", 2026.
5. Journal du Web, "GPT-5.6, OpenAI is already preparing its next flagship model" (in French), 2026.
6. Cryptopolitan, "GPT-5.6 rumors intensify as OpenAI eyes late June", 2026.
7. OpenAI, "GPT-5.6, frontier intelligence that scales with your ambition", July 9, 2026.
8. OpenAI, official API price list, accessed October 16, 2026.