GPT-5.6 is probably days away. Here is what the leaks confirm and what your bill might do.
A model ID leaked out of OpenAI's own Codex backend, the chief scientist reportedly called it a meaningful improvement, and the release-date market has it landing this week. The pricing? Nobody knows. But OpenAI's last three steps tell you more than the rumor mill does.
Photo by Brecht Corbeel on Unsplash
GPT-5.6 is real but unannounced. A gpt-5.6 tag surfaced in OpenAI's Codex backend in mid-May, The Information says the chief scientist called it a meaningful improvement over GPT-5.5, and the cadence plus a Polymarket reading near 85% point to a release the week of June 22. OpenAI has published no pricing, so GPT-5.5 at $5/$30 per million tokens is the only baseline you have. Our best guess: the sticker holds and OpenAI sells a lower effective cost through fewer tokens per task, the same trick it ran when 5.5 doubled the rate over 5.4. The rest of this post shows the math so you can judge the launch in minutes instead of refreshing X all day.
How much of this is real
Two signals are worth your attention. The first is concrete: a researcher spotted a gpt-5.6 routing identifier sitting next to gpt-5.5 in OpenAI's Codex rollout infrastructure in mid-May, then watched it get pulled. That is a name, not a spec sheet, but names do not show up in production backends by accident.
The second is The Information's report, around June 11, that Pachocki described GPT-5.6 internally as a meaningful improvement over 5.5. Secondhand, sure, a leaked Slack line rather than a press release. But The Information has a track record, and it lines up with the backend leak.
Everything past that is noise. The codenames floating around ("kindle-alpha" and a chain of others) contradict each other depending on which leak list you read, which is usually a tell that people are guessing. The 1.5M context window, the 10 to 15% efficiency gain, any benchmark number with a GPT-5.6 label next to it: invented. There is no model card to leak those from yet. Treat them as wish-casting until OpenAI ships a system card.
Why late June is the smart bet
OpenAI has settled into a roughly six to seven week rhythm on flagship point releases this year. Stack the dates up and the pattern does the forecasting for you.
| Model | Released | Gap from prior |
|---|---|---|
| GPT-5.4 | March 5, 2026 | - |
| GPT-5.5 | April 23, 2026 | ~7 weeks |
| GPT-5.6 (forecast) | June 22 to 28, 2026 | ~6 to 7 weeks |
The prediction markets agree. As of mid-June, Polymarket's "when will GPT-5.6 be released" market sat around 85% on the June 22 to 28 window, with most of the rest on it slipping past June 28. One caveat worth saying out loud: a closed beta does not count as a release for most developers. GPT-5.4 went to consumer ChatGPT first and reached the enterprise API weeks later. If you build on the API, your real access date may trail the headline launch.
The precedent that actually predicts your bill
Here is the part the rumor posts skip. When OpenAI went from GPT-5.4 to GPT-5.5, the headline price doubled. Input jumped from $2.50 to $5 per million tokens, and output went from $15 to $30.
| Model | Input / 1M | Output / 1M | vs prior |
|---|---|---|---|
| GPT-5.4 (Mar 2026) | $2.50 | $15.00 | baseline |
| GPT-5.5 (Apr 2026) | $5.00 | $30.00 | +100% |
A 2x hike sounds brutal, and OpenAI's counter was efficiency. It said GPT-5.5 finishes the same Codex task on roughly 40% fewer output tokens, so a doubled rate would not double your bill. The catch is that the claim did not survive independent testing. OpenRouter measured the token saving at closer to 19 to 34% on real long prompts, and The Register's analysis found that once you net everything out, GPT-5.5 still cost about 49 to 92% more per task than GPT-5.4. So the efficiency was real but partial. The price increase mostly stuck.
That is the lens for GPT-5.6. We have watched OpenAI raise the sticker and then defend it with efficiency math that overpromises, so a forecast that says "prices will fall" is fighting the most recent data point. The honest base case is a flat or higher sticker, an efficiency story attached to it, and a real bill that lands somewhere short of the marketing.
Three ways the pricing could land
None of these are confirmed. They are scenarios anchored to GPT-5.5's real numbers and OpenAI's recent behavior, so you can model your budget instead of waiting on a press release.
| Scenario | Input / 1M | Output / 1M | Effective vs 5.5 |
|---|---|---|---|
GPT-5.5 (confirmed baseline) | $5.00 | $30.00 | - |
Hold the line (most likely) Same sticker, efficiency does the work | $5.00 | $30.00 | ~10 to 15% lower if tokens drop |
Modest bump Signals an upgrade, efficiency offsets it | $6.00 | $36.00 | roughly flat per task |
Big-leap premium If OpenAI frames it as a major jump | $7 to 8 | $40 to 48 | higher |
Scenarios based on GPT-5.5 pricing and OpenAI's release history. No official GPT-5.6 pricing exists as of June 20, 2026.
The market it walks into
Whatever number OpenAI picks, it lands in a field that has gotten cheaper underneath it. Anthropic held Opus 4.8 at $5/$25 and added a faster tier. Google's Gemini 3.5 Flash undercuts everyone serious at $1.50/$9. DeepSeek made its 75% cut permanent. A $5 flagship is no longer the obvious default it was a year ago.
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| DeepSeek V4-Pro | $0.44 | $0.87 | 1M |
| Grok 4.3 | $1.25 | $2.50 | 1M |
| Gemini 3.5 Flash | $1.50 | $9.00 | 1M |
| Gemini 3.1 Pro | $2.00 | $12.00 | 1M |
| GPT-5.4 | $2.50 | $15.00 | 1M |
| Claude Opus 4.8 | $5.00 | $25.00 | 1M |
| GPT-5.5 | $5.00 | $30.00 | 1M |
| GPT-5.6 (forecast) | $5 to 8 | $30 to 48 | 1M+ |
The reported design goal for GPT-5.6 is long-running agentic work, multi-hour Codex sessions that stay coherent. If that holds, the pitch is not raw price, it is fewer babysitting interventions per job. That is a real value story, but it only pencils out if your tasks are long and your alternative keeps falling over. For a high-volume classification job, Gemini 3.5 Flash at a sixth of the output price is still the boring correct answer.
Know your own numbers first
If you are on GPT-5.5 today, sit tight. Even an on-time launch probably will not put GPT-5.6 in the enterprise API on day one, so there is nothing to migrate this week. The useful move is to know your own numbers before the pricing drops, so you can judge it in minutes instead of guessing.
Pull your real monthly input and output token volume. Run it against the scenarios above. If a flat $5/$30 with 10% better efficiency barely moves your bill, you can ignore the launch noise entirely. If a jump to $8 output would hurt, you want a cheaper fallback lined up now. Artificial Analysis tracks live quality-versus-price rankings if you want independent benchmarks the day GPT-5.6 appears. We keep every current API rate on the pricing comparison page.
The moment OpenAI publishes real GPT-5.6 pricing, we will update this post and add the model to the tables. Until then, use the cost calculator to see what each scenario would cost at your actual volume.
Sources
- - Codex backend leak analysis, DEV.to and AIxploria, June 2026
- - The Information on Jakub Pachocki's internal note, reported June 11, 2026
- - Polymarket: "When will GPT-5.6 be released" market, read mid-June 2026
- - OpenRouter and The Register on GPT-5.5 real-world cost vs GPT-5.4, May 2026
- - OpenAI developer docs, GPT-5.5 and GPT-5.4 model pages
- - TokenCost pricing data for current frontier models