Alphabet had a genuinely good quarter. Revenue hit $119.8 billion, up 24% year-over-year. Google Cloud revenue grew 82% to $24.77 billion. Net income landed at $112.1 billion. Gemini now processes 22 billion API tokens per minute, the Gemini app claims 950 million monthly active users, and Gemini Enterprise has reached nearly 90% of the Fortune 100. By almost any normal measure of a public company’s health, this was a strong report.
And yet the most consequential sentence in Wednesday’s call wasn’t a revenue number. It was Sundar Pichai, asked directly about Gemini 3.5 Pro’s now-repeated delay, saying that “coding and agentic coding is an example of that” — an area where Gemini still needs to improve.
That’s not a hostile analyst’s framing. That’s the CEO, on the record, on the one call every quarter where a misstatement gets transcribed, fact-checked, and priced into the stock.
The admission behind the delay#
Gemini 3.5 Pro was originally expected in June. It missed June. It missed a mid-July target. As of this week it’s still described as being tested with partners, with DeepMind’s Logan Kilpatrick offering only that Google “hopes to land soon.” Bloomberg reported on July 16 that the underlying cause wasn’t a scheduling slip at all — Google ran a late-June retrain aimed specifically at coding performance, and the internal results were, in Bloomberg’s word, “disappointing.” Alphabet shares fell as much as 4.4% on that report, erasing roughly $200 billion in market cap in a single session.
The earnings call is the first time a Google executive has connected those dots publicly rather than leaving them to anonymously-sourced reporting. Investors pushed on it because, as Calcalist’s coverage of the call put it, Gemini 3.5 Pro “was seen as critical to strengthening the company’s position in AI coding and autonomous ‘agentic’ systems” — exactly the category where Anthropic has spent 2026 building a commercial moat.
Pichai’s response was a pivot, not a defense. Rather than re-commit to a Gemini 3.5 Pro date, he redirected to Gemini 3.6 Flash — the Flash-tier model Google shipped on July 21, one day before the call — citing its coding-benchmark gains and lower token cost. He also teased that Gemini 4 models would arrive on “almost a monthly cadence,” a roadmap promise that, notably, commits to nothing about coding capability specifically. He closed with the standard-issue line every CEO reaches for in this spot: Alphabet remains “very committed and very confident” about its AI leadership position.
Flash gains are real. They’re also not the point.#
To be fair to Google, the Flash-tier numbers aren’t nothing. Gemini 3.6 Flash posted a DeepSWE score of 49%, up from 37% on the prior version, and Gemini 3.5 Flash-Lite jumped from 31% to 54% on Terminal-Bench 2.1 — even edging out the larger Gemini 3 Flash on SWE-bench Pro (54.2% versus 49.6%). Those are genuine, publishable improvements, and we covered them as such the day they shipped.
But Flash was never the model anyone was waiting on. Flash is Google’s cost-efficient workhorse tier — useful, cheap, fast, and explicitly not the frontier coding model the market has been asking about since May. Citing Flash benchmarks in response to a question about Pro’s delay is a category substitution: it answers “is Google shipping anything” with real data, while leaving “can Google’s best model actually code” unanswered for the fourth consecutive month.
Independent verification firm Artificial Analysis had already flagged the deeper issue a day before the earnings call: despite the real coding-benchmark gains, 3.6 Flash didn’t actually move Google’s own aggregate Intelligence Index at all. Coding-specific improvement and general model quality aren’t moving together — which is a plausible explanation for why a coding-focused retrain of the much larger Pro model produced results Google itself wasn’t ready to ship.
The cadence gap, in the same 24 hours#
Here’s the part that makes this an Anthropic story as much as a Google one. While Pichai was teasing an “almost monthly” release cadence for a model family that doesn’t exist yet, Claude Code shipped two dated, changelog-documented point releases back to back:
v2.1.217 (July 21) added emoji autocomplete, fixed a memory leak in truncated MCP tool output, patched a background-session symlink escape, capped concurrent subagent spawning at 20 by default (CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS), and fixed --max-budget-usd not actually stopping background subagents once the cap was hit.
v2.1.218 (July 22 — the same day as the earnings call) moved /code-review to run as a background subagent so review work stops filling the main conversation, added HTTP error surfacing for failed MCP connections, fixed /ultrareview silently mishandling descriptive arguments, and closed out more than twenty other dated bug fixes, from Windows path corruption to screen-reader announcement timing.
Neither release is a headline-grabbing model drop. That’s exactly the point. This is what a mature, shipping-cadence-as-strategy operation looks like: dozens of small, dated, independently verifiable fixes landing daily, against a backdrop of Anthropic’s own model cadence — Sonnet 5 on June 30, Fable 5 restored to general availability July 1, Opus 4.8 in general release since May 28 — with no retrain drama and no “hopes to land soon” quotes required.
Why this matters beyond one earnings call#
Google is not behind on AI generally. A 950-million-user Gemini app, 90% Fortune 100 penetration for Gemini Enterprise, and 82% cloud growth are not the numbers of a company in trouble. But “AI generally” and “AI for agentic coding” are increasingly different markets with different competitive dynamics, and Wednesday’s call is the clearest signal yet that Google knows it’s behind in the second one specifically — behind enough that its own CEO named the gap out loud rather than let another quarter of silence do the talking.
For engineering teams choosing a coding stack today, the practical read is simple: Alphabet’s own leadership isn’t asking you to wait for Gemini 3.5 Pro. It’s asking you to trust that Gemini 4 will arrive “almost monthly” and fix what 3.5 Pro couldn’t. Claude Code isn’t asking for that kind of faith — it shipped twice while Alphabet was still on the call explaining why its flagship coding model hasn’t shipped once.
Sources#
- Alphabet Q2 2026 earnings summary — 9to5Google
- Alphabet Q2 2026 earnings call transcript — Benzinga
- Pichai’s response on Gemini coding delay — Calcalist
- Google Q2 2026 earnings live coverage — CNBC
- Gemini 3.5 Pro retrain delay reporting — Bloomberg, July 16, 2026 (previously covered)
- Claude Code changelog — code.claude.com