Leak: OpenAI Pushed GPT-5.6 to July as DeepMind Holds Gemini 3.5 Pro
The claims are unconfirmed, but they land as official docs show no GPT-5.6 listing and Google has already missed its June target for 3.5 Pro.
By Ryan Merket ยท Published
Why it matters
The alleged delays show how model cadence has become a trust signal: labs are better off missing a window than shipping frontier models that fail developers, benchmarks or safety gates.
A pseudonymous X account, leo (@synthwavedd), said in a two-post thread on X on Tuesday that OpenAI's GPT-5.6 release has slipped from the week of June 23 to roughly mid-July, while Google DeepMind is holding Gemini 3.5 Pro past its June launch window because the lab is not satisfied with the model's current state.
The claims are not confirmed by OpenAI or Google DeepMind. That matters because both companies have spent the past year turning model cadence into a competitive signal: release windows, even soft ones, now tell developers when to migrate workloads, investors where the perceived frontier sits, and enterprise buyers which vendor is likely to set the next benchmark cycle.
The X thread says GPT-5.6 "has been delayed and will no longer release this week" and gives a new target of "~mid-July." It also says DeepMind is not satisfied with "the current state of 3.5 Pro" and that the model will no longer launch in June 2026. The captured thread further names "Bidi" as OpenAI's new voice model, but the source text cuts off before stating the status of those preparations.
OpenAI's public record does not currently confirm GPT-5.6. The company's April 23 GPT-5.5 launch post describes GPT-5.5 as rolling out to ChatGPT and Codex, with an April 24 update saying GPT-5.5 and GPT-5.5 Pro were available in the API. OpenAI's ChatGPT model help page lists GPT-5.5 Instant, Thinking and Pro, but a search of the page and the broader OpenAI release notes did not show a GPT-5.6 entry. That does not disprove an unreleased model, but it means the alleged delay is still a leak, not a public schedule change.
The DeepMind side has a firmer public timeline. In a May 19 Google blog post, Google DeepMind executives Koray Kavukcuoglu, Jeff Dean, Oriol Vinyals and Noam Shazeer introduced Gemini 3.5 Flash and said the team was "hard at work" on Gemini 3.5 Pro, which was already being used internally and expected to roll out "next month." On that timeline, Gemini 3.5 Pro was supposed to arrive in June 2026. As of Tuesday, June 23, DeepMind's public model card index lists Gemini 3.5 Flash and Gemini 3.5 Audio, but not Gemini 3.5 Pro.
That gap is the news. Google did not merely have a vague product in the lab; it put 3.5 Pro on a public clock. The May 19 post said 3.5 Flash was already generally available through the Gemini app, AI Mode in Search, Google Antigravity, the Gemini API in Google AI Studio and Android Studio, Gemini Enterprise Agent Platform and Gemini Enterprise. It positioned Flash as the speed and agentic-workflow member of the family, while reserving Pro for the next step up. If the leak is right, DeepMind is choosing to absorb a missed June window rather than ship a frontier model it does not think clears the bar.
That choice would fit the new economics of model releases. Frontier labs can no longer treat a model launch as a one-day marketing event. A rushed release now cascades through consumer chatbots, coding agents, API customers, enterprise pilots, safety evals and benchmark comparisons. A weak Pro model would not only disappoint Gemini users; it would also give OpenAI, Anthropic and xAI a clean comparison point at the moment Google has been trying to make Gemini feel less like a search-adjacent product and more like an agent platform.
OpenAI faces a different pressure. GPT-5.5 is already the public baseline. In its April announcement, OpenAI said GPT-5.5 was designed for complex work across coding, online research, data analysis, documents, spreadsheets and software operation. It also priced the model above GPT-5.4 in the API while arguing that token efficiency offset some of that cost. A GPT-5.6 release, if real, would have to justify itself against that promise rather than simply beat an older model on a leaderboard.
The voice-model thread is thinner, but still strategically relevant. OpenAI's current developer docs list GPT-Realtime-1.5 as its flagship audio model for voice agents and customer support, with text, audio and image input and text and audio output. OpenAI has also been building toward more programmable speech systems: its March 2025 audio-model release introduced newer speech-to-text and text-to-speech models and emphasized steerable voice generation. A new model nicknamed Bidi would sit in that same lane, but the X thread does not provide enough complete text to report a launch date or capability claim.
The safest reading is narrow: a widely viewed leak account says two major model launches have slipped, and the public documentation is consistent with there being no GPT-5.6 or Gemini 3.5 Pro launch as of June 23. The stronger conclusion is about timing. Both OpenAI and Google DeepMind are managing a market that expects rapid, visible iteration, but the next round of frontier models is being judged less on novelty than on whether it improves agentic work, coding reliability, voice interaction and enterprise-grade safety without creating new cleanup work for customers.
If the mid-July target for GPT-5.6 holds, OpenAI would be spacing the release roughly three months after GPT-5.5. If Gemini 3.5 Pro slips out of June, Google will have to explain the delay through the product itself: a late model can be forgiven if it moves the frontier, but not if it arrives as a patched version of a public promise.