Google has missed its own deadline for Gemini 3.5 Pro for a third time, and this time the company isn’t hiding the reason: the model still isn’t good enough at coding to ship. Bloomberg reported on July 16 that Google DeepMind’s flagship reasoning model has fallen short of internal quality bars, pushing the release past the June target promised at I/O 2026 and now past a quiet mid-July target as well — a delay first detailed by 9to5Google.
The company confirmed to Bloomberg that it’s “taking time to improve” the model’s capabilities, “particularly in coding,” and that engineers updated Gemini’s training data in late June specifically to fix that weakness. The results, according to the report, were disappointing enough that the fix didn’t stick.
What actually happened
Here’s the timeline as it stands: Gemini 3.5 Flash shipped on schedule at I/O in May, with Google promising the Pro variant would follow in June. June came and went with no launch. We covered the next chapter of that story on decodegear — Google didn’t just slip the date, it scrapped the original base model and rebuilt it from scratch after concluding the coding and tool-calling failures were structural, not the kind of thing a fine-tuning pass fixes. That rebuild was supposed to land by mid-July. It hasn’t.
Google’s public line is unchanged: the company is “currently testing 3.5 Pro, an upgraded Flash model, and other models with partners,” and reiterating a commitment to “shipping quickly.” Notice what’s missing from that statement — an actual date.
Why Google keeps stumbling on coding
Coding has become the benchmark that actually moves the needle for enterprise AI buyers, and it’s also the hardest thing to fake. A model can sound fluent and still generate code that fails silently — wrong edge cases, broken recursive tool calls, malformed structured output. Reporting around this delay points to exactly those failure modes: multi-step reasoning chains that lose the thread, and tool-calling behavior that doesn’t hold up outside of curated demos. That’s a different class of problem than “the model needs more time to cook” — it’s the kind of gap that shows up the moment real developers start hammering on it in production, which is presumably why Google keeps pulling it back rather than shipping something that embarrasses itself on day one.
The competitive backdrop makes the delay sting more. GPT-5.6 has set a bar on coding benchmarks that Gemini 3.5 Pro is reportedly still chasing, and Anthropic and other labs haven’t slowed down while Google works through its rebuild. We’ve been tracking that broader scramble — including smaller labs like Moonshot AI trying to close the gap with cheaper, faster models — and every extra week Gemini 3.5 Pro stays unreleased is a week competitors get to look like the safer bet for teams picking a coding assistant right now.
What’s next for Gemini 3.5 Pro
Google hasn’t given a new hard date, which is itself a signal after burning through two previous ones. Prediction markets tracking the launch have shifted their consensus toward late July or early August rather than the original summer window, though that’s crowd betting, not a company announcement, and should be read as exactly that.
For everyday users, the practical impact is limited — Gemini 3.5 Flash is already live and handles most everyday tasks fine. The people actually affected are developers and enterprise teams who were counting on Pro-tier coding performance for real workloads, and who are now deciding whether to keep waiting or route that work to a competing model in the meantime.
We’ll keep updating this story as Google confirms a real ship date. Until then, treat any specific “launch day” claim you see online as speculation — Google itself hasn’t committed to one.
Should you wait for it?
If you’re picking an AI assistant for coding work today, the honest answer is: don’t plan around a Gemini 3.5 Pro release date, because Google hasn’t given you one to plan around. Two missed deadlines and a from-scratch rebuild is a pattern, not a hiccup — the kind of pattern that tends to produce a fourth slip before it produces a launch. Teams with real deadlines of their own are better served picking whichever model is actually shipping and available right now, and revisiting the decision once Gemini 3.5 Pro is out in the wild long enough for independent benchmarks to confirm it holds up outside Google’s own testing.
That’s also the more interesting story here than the delay itself: a model missing a date is routine. A model getting pulled back to the studs after its base was already built is Google admitting the first version wasn’t fixable with patches — and that’s a much bigger tell about how far off Gemini 3.5 Pro actually was.
Deixe um comentário