Google / Gemini 4
Pichai: Gemini's Next Leap Needs Bigger Base Models
At Alphabet's Q2 2026 earnings call on July 22, 2026, Google CEO Sundar Pichai said the key to the next Gemini leap is "much larger base models." The comment landed just one day after Gemini 3.6 Flash shipped — here's what that timing means.
Why It Matters
"The next leap needs much larger base models"
May's Gemini 3.5, then Gemini 3.6 Flash just a day earlier. The comment came right on the heels of back-to-back launches — from the CEO himself.
On July 22, 2026, during Alphabet's Q2 2026 earnings call, Google CEO Sundar Pichai said Gemini's next leap "depends on building much larger base models." According to reporting from THE DECODER, he went on to say "We have started our most ambitious pre-training run yet for Gemini 4, and are excited by the progress we are seeing at the frontier."
The timing is no coincidence. Just one day before the earnings call, on July 21, Google shipped two new models: Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. According to 9to5Google's coverage, 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens, and uses roughly 17% fewer output tokens than its predecessor on the Artificial Analysis Index. Yet what Pichai chose to emphasize wasn't efficiency — it was scale. That gap is exactly what makes the remark newsworthy.
The debate over scaling laws — the premise that bigger models keep getting better — has split the industry since late 2025. Some researchers at OpenAI and Anthropic have shifted weight toward test-time compute and reinforcement-learning post-training, arguing that pretraining scale alone is starting to plateau. Against that backdrop, one of the industry's biggest players flatly declaring that "the next leap still needs bigger base models" reads as a direct counterpoint in the scaling-law argument, and a signal worth watching for where the industry heads next.
Follow the Money
Bigger base models come with a bigger price tag
Numbers disclosed at the same earnings call give Pichai's words real weight.
Google raised its full-year capex outlook from $180–190B to $195–205B. Quarterly capex hit a record $44.9B, nearly double the same period a year earlier. "Much larger base models" translates directly into "much larger compute and cost." The scaling race is as much a fight over capital allocation and accountability to shareholders as it is a technical argument.
Who It Affects
What this comment means, by reader type
The same sentence lands differently depending on where you sit.
Business
Gemini 4's arrival remains hard to predict. That's not a reason to rush switching off your current Gemini or rival API contract, but it is worth using the Gemini 4 timeline as leverage when negotiating your next contract renewal.
PM
The Flash lineup (3.6 Flash / 3.5 Flash-Lite) is already usable today. A two-track roadmap — build features on the lightweight models now, refresh with Gemini 4 later — is the realistic move. Pichai hinted at a near-monthly release cadence going forward, so plan an evaluation process built for frequent swaps.
Engineer
Google itself has admitted it's behind on coding and agentic coding. Pichai said "there are many attributes on which we are still at the frontier. There are areas where we've acknowledged we need to improve; coding and agentic coding is an example of that." Gemini 3.5 Pro, originally targeted for June, slipped to a July 17 target and was still unreleased as of the earnings call. It's safer to design coding workflows assuming you'll run multiple models side by side.
What's Next
What to watch from here
Gemini 4's launch timing
No release date confirmed. Pichai hinted at a "near-monthly" cadence going forward but stopped at calling the pretraining run "most ambitious," without naming a public launch window.
Gemini 3.5 Pro's fate
Repeated delays trace back to a rebuild of coding performance. Whether Gemini 3.5 Pro or Gemini 4 ships next will reveal a lot about how Google's roadmap is actually playing out.
Recommended actions
1) Don't rush a decision to switch to a bigger model — wait and watch. 2) Adopt the Flash line's efficiency gains now. 3) Design coding workloads to run on multiple models in parallel.
The Other Side
Optimism alone doesn't tell the whole story
The bet that "bigger wins" still has plenty of skeptics.
Betting everything on scale carries real risk. First, cost: as capex balloons to $195–205B, investors are increasingly asking whether that spending will actually pay off. Second, a fork in approach: some researchers at OpenAI and Anthropic, along with rivals like DeepSeek, have shifted focus away from pretraining scale toward test-time compute, reinforcement learning, and data quality — "bigger is better" is not the only strategy on the table. Gemini 3.5 Pro's repeated delays are themselves a reminder that scale alone doesn't solve every problem, coding accuracy among them.
| Betting on scale (Google and some peers) | Betting on a different axis (test-time compute, efficiency) |
|---|---|
| Grow the pretrained base model | Prioritize reinforcement learning and test-time compute |
| Raise capex to $195–205B | Chase equal performance with less compute |
| Ship models at a near-monthly cadence | Differentiate with specialized, lightweight models |
"We have started our most ambitious pre-training run yet for Gemini 4." — Sundar Pichai
Bottom Line
Read it in dollars, not just words
On its own, Pichai's comment sounds like the usual "the next model will be great" teaser. Read alongside the same day's upward revision to capex guidance, though, it's both a technical statement of intent and an explanation to shareholders that Google is still betting on scale. That this comment landed on July 22, right after back-to-back launches — Gemini 3.5 Flash in May and 3.6 Flash on July 21 — shows Google's hand: the next round is still being fought on model size. Full financial detail is available in CNBC's earnings coverage.