Alphabet CEO Sundar Pichai stated during the company’s Q2 2026 earnings call that Google needs to enhance its coding and agentic coding capabilities. He emphasized that a larger Gemini 4 base model is essential for maintaining competitiveness in the evolving AI landscape.
These comments followed the introduction of Gemini 3.6 Flash, which Google released one day prior. Pichai noted that Gemini 4 is currently in pretraining and acknowledged that the unreleased Gemini 3.5 Pro faces delays due to coding issues.
Pichai asserted confidence in Google’s strengths but admitted that there is room for improvement in coding and agentic coding. He described the launch of 3.6 Flash as a positive development, stressing the importance of larger base models for future breakthroughs. Previously, in May, during the Hard Fork podcast, he indicated that Google was “a bit behind” in agentic coding because it lacks a developer-facing product that generates valuable usage data, which competitors are already leveraging.
As for the much-anticipated flagship model, Gemini 3.5 Pro has yet to be released. Google had announced the 3.5 series at the I/O event in May, launching 3.5 Flash simultaneously, with plans for 3.5 Pro to debut the following month. Google now indicates that it is “currently testing with partners,” with a broader release contingent on readiness. Bloomberg reported that coding performance issues contributed to the delay, as noted by SEJ. An update to the model’s training data aimed at improving its coding capabilities did not yield expected results.
The delays have caused unrest within the company, leading to the departure of two senior AI researchers in June: Noam Shazeer, co-lead of Gemini, and John Jumper from AlphaFold, who left for OpenAI and Anthropic, respectively, amid concerns regarding Google’s standing in AI coding tools.
Recently, Google launched Gemini 3.6 Flash, an updated version providing improvements in coding and efficiency. Notably, this version produces 17% fewer output tokens than the previous 3.5 Flash, enhancing cost-effectiveness while bolstering coding abilities.
The current status of the Gemini models is summarized as follows: Gemini 3.5 Flash is available as a general-purpose model, while Gemini 3.6 Flash, also available, serves as an updated workhorse. Gemini 3.5 Flash-Lite has been released as a faster, lower-cost alternative, whereas Gemini 3.5 Pro is still in partner testing with no confirmed release date. Gemini 4 remains in pretraining without a confirmed timeline for its launch.
On the DeepSWE coding benchmark, Google reported a score of 49%, an increase from 37% with 3.5 Flash. Google characterizes 3.6 Flash as the new standard workhorse, while both 3.5 Flash and the newly introduced Flash-Lite tier are being integrated into Google Search.
Moving forward, the critical focus lies on the timely release of 3.5 Pro and adherence to the monthly release schedule mentioned by Pichai. Successfully meeting these targets will demonstrate Google’s capability to transition from planning to delivering new products. Gemini 4 is seen as a long-term goal, with its pretraining process described as the company’s most ambitious effort yet, though no release date has been set.
