Google has once more pushed back Gemini 3.5 Pro, its flagship model that the whole industry expected on July 17. The company reportedly scrapped the near-finished version to restart on a fresh architecture, after results judged disappointing on coding and complex reasoning. Alphabet shares fell about 4% on the reports.
Key Takeaways
- Gemini 3.5 Pro slips again, after a June promise and then a July 17 target
- Google reportedly dropped the near-final model to restart the architecture from scratch
- No model card, no pricing page, no official benchmark accompanies the announcement
Have an AI Sum Up This Article
ChatGPTAnother Delay, This Time a Rebuild
Gemini 3.5 Pro’s timeline has stretched since Google I/O 2026. The company showed Gemini 3.5 Flash at its conference, then promised the Pro version for June, a deadline already missed. The target then shifted to July 17, with no public confirmation.
This new delay looks less like a simple slip. According to the reports, Google chose to throw out the near-ready version and restart from a fresh base. A heavy call, one that pushes back availability of the most anticipated model in its lineup.
Two failure points keep coming up. The scrapped model reportedly could not hold structural consistency when generating complex, multi-layered SVG scenes, and broke down inside recursive tool-calling environments. Two capabilities squarely expected of a model meant to compete at the top, as our own Claude, ChatGPT and Gemini test for pro use already showed.
In late June, Google updated its training data to sharpen coding. The results reportedly disappointed, which tipped the decision toward a rebuild rather than a patch. The silence on specs, including the two-million-token context window floated earlier, remains total.
The delay runs deeper than a scheduling slip. The model is reportedly months behind its original internal plan, a rare gap for a product Google placed at the heart of its answer to Anthropic and OpenAI. Restarting at this stage admits the previous version was no longer catching the front of the pack.
User Impact: A Hole in the Lineup
For developers and product teams, the missing Pro leaves a gap. Gemini 3.5 Flash exists, but it targets speed and cost, not heavy reasoning loads or demanding code. Those who waited for the step up stay on the previous version or move elsewhere. Google keeps shipping other Gemini bricks, Nano Banana offering free images.
The pricing fog makes the wait worse. With no public grid and no model card, no team can plan a migration or cost out a move to scale. The gap between announcement and real availability, already flagged on other launches, becomes a direct brake on decisions here.
The contrast is sharp with models already in production. When OpenAI opened the worldwide rollout of GPT-5.6, developers got a stable, documented, priced target. Facing them, Gemini 3.5 Pro stays a promise with no firm date.
The targeted weakness, coding, hits the most contested ground of the moment. That is where real usage tests play out, and that is where the delay costs the most on adoption.
For teams already committed to Gemini, the wait forces a trade-off. Staying on the current generation freezes their capabilities, while moving to a rival costs integration time. Neither call is easy to make while Google publishes no firm date and no scope for the coming Pro version.
More articles on Horizon
- Kimi K3 Closes the Gap With Top US AI Models
- Anthropic Cuts Fable 5 Limits Starting July 20
- GPT-5.6 Deletes User Files in Full Access Mode
Rival Impact: Anthropic and OpenAI Pull Ahead
The delay lands at the worst competitive moment. Anthropic and OpenAI are shipping models that top Gemini’s current capabilities on several tasks. Every week of slippage widens the perceived gap at the top of the market.
Inside Google, the mood is tense. Engineers, researchers and managers reportedly voiced frustration, with a fear of losing an edge on ground the company dominated not long ago. The rebuild is framed as a bet on quality, but it also carries a calendar cost that the share drop made public. On a field where reputation compounds with every shipped model, ceding even a few weeks at the top is harder to claw back than it looks.
Pricing pressure fills out the picture. On code, rivals are already cutting prices, as with Grok 4.5 undercutting Claude Opus at a third of the price. A missing Gemini Pro lets that segment take shape without it.
Our read: the bet can hold if the rebuild delivers a model truly at the level. But the market does not reward promised spec sheets, it rewards models that run. As long as Pro has no date, Google defends its spot with Flash and an aging version.
The calendar stays the real judge. A promised model captures neither market share nor developers, it hands the ground to models that actually ship. Every week without a firm Pro date turns the targeted technical lead into a concrete commercial risk.
Follow the story on Horizon.
