During Alphabet Q2 2026 Earnings CallCEO Sundar Pichai said Google needed to improve its coding and agent coding capabilities, and that a larger Gemini 4 base model would be needed to remain competitive in the next frontier.
These comments came a day after Google introduces Gemini 3.6 Flash and announced that Gemini 4 is currently in pre-training. Meanwhile, the unreleased Gemini 3.5 Pro remains delayed due to coding issues.
What Pichai said
When asked if Gemini could stay on the frontier, Pichai said he was confident that Google still excels in many areas, but acknowledged that coding and agent coding require further development.
He highlighted the recent launch of 3.6 Flash as a positive step, while outlining future plans, explaining that the next advancement depends on larger base models. Google is currently training Gemini 4, which Pichai said Google will need to compete at that level.
This is consistent with his previous comments. In May, Pichai mentioned on the Hard Fork podcast that Google was “a little late” in agent coding due to the lack of a developer-facing product that generates valuable usage data collected by competitors.
The flagship product is still not released
Google’s Gemini 3.5 Pro hasn’t shipped yet. The company announced the 3.5 series at I/O in May, releasing version 3.5 Flash the same day and announcing that version 3.5 Pro would follow the following month.
Google now says 3.5 Pro is “currently in testing with partners” and will be widely available “as soon as it’s ready.” Bloomberg reported this month that coding performance contributed to the delay, as reported by SEJ. A late June update to the model’s training data, intended to improve coding, fell short of expectations.
The pressure manifested itself elsewhere. In June, two Google AI principal researchers, Noam Shazeer, co-lead of Gemini, and John Jumper of AlphaFold, left for OpenAI and Anthropicamid concerns within DeepMind about Google’s position in AI coding tools.
What Google Shipped
Google recently released Gemini 3.6 Flashan update to its reliable Flash level rather than a flagship model. The company states that 3.6 Flash produces 17% fewer output tokens compared to 3.5 Flash and is more cost effective, with improved encoding capabilities.
| Model | Current status | Intended role | Timing |
|---|---|---|---|
| Gemini 3.5 Flash | Available | General Purpose Flash Model | Released at Google I/O |
| Gemini 3.6 Flash | Available | Updated workhorse with coding and efficiency improvements | Released July 21 |
| Gemini 3.5 Flash-Lite | Available | Faster, lower cost model for high-volume workloads | Released July 21 |
| Gemini 3.5 Pro | Testing with partners | Flagship model of the 3.5 series | No release date confirmed |
| Gemini 4 | In pre-training | Next-generation base model | No release date confirmed |
On DeepSWEa coding benchmark cited by Google, it scored 49%, compared to 37% with 3.5 Flash. Google describes 3.6 Flash as the new standard workhorse, while 3.5 Flash remains available. They also introduced a more affordable Flash-Lite 3.5 tier, now integrated with Google Search.
Why it matters
The most important thing is the timeline rather than how Google compares to others. They’ve already introduced Flash models in AI and Search mode, but version 3.5 Pro slipped and Gemini 4 doesn’t have an official date yet.
Looking to the future
What really matters now is whether version 3.5 Pro will finally be available and whether Google sticks to the monthly release schedule Pichai mentioned. Successfully achieving these goals would show how well Google can turn its projects into actual, shipped products.
On the other hand, Gemini 4 is more of a long-term goal. Google considers this pre-training process to be its most ambitious yet, but has not yet set a release date for the model.
Featured Image: ID Thrive Studios/Shutterstock
