Gemini 3.5 Pro Still Missing at 67 Days, Rivals Gain [2026]

Google promised Gemini 3.5 Pro would ship in June 2026. Then it promised July 17. Both dates have now come and gone, and as of July 25, 2026, Alphabet’s flagship large language model still has no public release date, no listed entry in the Gemini API, and no official specification sheet.

That is 67 days since Google chief executive Sundar Pichai stood on stage at Google I/O on May 19, 2026, and told a packed developer audience to “give us until next month” for Gemini 3.5 Pro. Google’s own API changelog shows it never hit that implied June target. A second, leaked target of July 17 also passed without a launch, and on July 21 Reuters reported that Google shipped three cheaper, lighter Gemini models instead, with no new timeline offered for the flagship Pro.

For Australian developers and enterprises weighing Google Cloud’s Vertex AI against Anthropic’s Claude, OpenAI’s GPT-5.6 and xAI’s Grok, the delay is no longer a rounding error. It is becoming the defining story of Google’s AI year.

Google’s Flagship AI Model Is Now 67 Days Late

The timeline is straightforward, even if Google’s public messaging about it has not been. At I/O on May 19, 2026, Pichai unveiled Gemini 3.5 Pro on stage but stopped short of shipping it, telling attendees the wait would not be long. That drew groans from the developer crowd, according to Business Insider’s coverage of the keynote.

June came and went with no launch. By late June, Google’s own Gemini API changelog showed no general-availability entry for the model, confirming the first deadline had slipped. Tech-insider.org covered that slip on July 1, when Alphabet shares fell amid the delay news.

A second date began circulating in mid-July: July 17, 2026, reported by TechTimes on July 13 as a leaked internal target, alongside unconfirmed claims about a 2-million-token context window and a “Deep Think” reasoning mode. Tech-insider.org covered that leaked target on July 17 itself, the same day reports emerged that four senior Google researchers had left for Anthropic.

July 17 arrived. Nothing shipped. “As of July 17, there’s still no sign of the model,” Mashable reported on July 24, describing Gemini 3.5 Pro as the case of the missing AI model. Today, eight days past that second deadline and 67 days past the original promise, Gemini 3.5 Pro remains unreleased.

What Google Shipped Instead on July 21

Rather than the flagship model, Google used the week after the July 17 target to ship something else entirely. Reuters reported on July 21 that Google released three cheaper, lighter-weight Gemini models, while explicitly declining to give any timing update for Gemini 3.5 Pro, which the outlet described as already delayed by several weeks.

It is a familiar pattern in software: when the headline feature is not ready, ship the supporting cast instead. The move gives Google something to announce, keeps the Gemini product line looking active, and buys additional time for the Pro model’s training or safety-review process, without directly addressing why the flagship keeps slipping.

The optics are not ideal. Google’s own product pages tell the same story. A review by AI workflow company eesel.ai, published July 21, found that the public Gemini Pro model page still listed Gemini 3.1 Pro as the current flagship, with a “3.5 Pro coming soon” badge sitting above it. The Gemini API’s public model list, the review found, had no gemini-3.5-pro entry anywhere.

That matters because Google has, at various points, said the model was being tested with partners. Reuters’ July 21 report confirmed as much, noting Google said Gemini 3.5 Pro was being tested with partners and “coming soon” — the same phrase Google has used since at least early July, according to multiple outlets tracking the release.

Timeline: Every Missed Deadline So Far

Stripped of speculation, the confirmed public timeline for Gemini 3.5 Pro looks like this:

Date	Milestone	Outcome
May 19, 2026	Google I/O: Pichai says Gemini 3.5 Pro is coming “next month”	Implied June target set
Late June 2026	Original June target window closes	Missed — no GA entry in Gemini API changelog
July 1, 2026	Google’s slip to July confirmed; Alphabet shares fall	First delay publicly reported
July 13, 2026	Leaked reports name July 17 as the new target	Unconfirmed by Google
July 17, 2026	Leaked target date arrives	Missed — no launch, no announcement
July 21, 2026	Google ships three cheaper Gemini models instead; Reuters reports no new timeline for Pro	Flagship still delayed
July 25, 2026	67 days since the original I/O promise	Still no gemini-3.5-pro listing in the public API

Two things stand out. Every confirmed date in this sequence is either a promise or a miss — there is no confirmed shipping date anywhere in the timeline. And the gap between the original promise and today keeps growing, with Google yet to publish a replacement date to reset expectations against.

What Developers See When They Check the API

For Australian developers building against the Gemini API, the delay is not an abstraction — it shows up directly in tooling. Querying Google’s public model-listing endpoint for the Gemini API returns exactly the models Google has actually shipped, and nothing else:

curl "https://generativelanguage.googleapis.com/v1beta/models?key=$GEMINI_API_KEY"    | grep -o '"name": "models/gemini[^"]*"'  # Output as of July 21, 2026 (per eesel.ai's review): # "name": "models/gemini-3.1-pro" # "name": "models/gemini-3.5-flash" # ... no gemini-3.5-pro entry anywhere in the list

As of that most recent review, the query returns Gemini 3.1 Pro and Gemini 3.5 Flash, but no gemini-3.5-pro identifier of any kind. Developers who built integration code around the leaked July 17 date, expecting to swap in a new model identifier, have nothing to point their SDKs at.

That is a meaningfully different situation from a model that is merely rate-limited or restricted to an allowlist, as OpenAI’s GPT-5.6 variants currently are. Gemini 3.5 Pro is not throttled access — it is unshipped, full stop, at least as far as any publicly reachable endpoint is concerned.

The Specs Nobody Can Officially Confirm

Much of what has been written about Gemini 3.5 Pro is, by every outlet’s own admission, unconfirmed. TechTimes was explicit about this in its July 13 report, noting that every specific claim circulating about the launch — including the date itself, the context window, and the benchmark numbers — comes from third-party reporting and unnamed insiders.

The specs repeated most often across coverage: a 2-million-token context window, a built-in “Deep Think” reasoning mode, and API pricing that different outlets have estimated anywhere from roughly $1.25 per million input tokens and $10 per million output tokens on the low end, to as much as $15 and $60 per million tokens on more speculative high-end estimates. None of these figures appear on an official Google pricing page, because Gemini 3.5 Pro does not have one yet.

Benchmark comparisons against GPT-5.6, Claude Sonnet 5, or Claude Opus 4.8 are similarly absent from any official source. Third-party leaderboard numbers have circulated, but without a model card or reproducible test conditions published by Google, none of them can be verified — which is why this article does not repeat specific benchmark scores for a model Google has not shipped.

The honest summary, a week past the second missed deadline: Google has confirmed that Gemini 3.5 Pro exists and is in testing. It has not confirmed what it costs, how large its context window actually is, or how it performs.

Why the Delay Keeps Extending

Google has not published a detailed explanation for either missed deadline. The closest thing to an official account is Reuters’ framing that the model is still being tested with partners — language that suggests an extended evaluation or safety-review phase rather than a simple technical failure.

Secondary reporting has offered more specific, harder-to-verify theories, including claims that Google DeepMind altered its training approach mid-cycle, pushing the schedule back further. Tech-insider.org has not been able to independently verify that claim and treats it as reported rather than confirmed.

What is better documented is the timing overlap with a separate story: the departure of four senior Google researchers to Anthropic, first reported around the July 17 leaked target date. Tech-insider.org covered that development in detail at the time. A single staffing story does not explain a training delay on its own, but the coincidence has fuelled speculation that Google’s flagship AI team is under unusual internal pressure during exactly the window it needed to hit two consecutive deadlines.

Whatever the internal cause, the external effect is consistent: two deadlines, two misses, and a growing gap between what Google promised in May and what it has shipped by late July.

How Anthropic Used the Gap: Claude Sonnet 5 and Fable 5’s Return

While Gemini 3.5 Pro sat in limited testing, Anthropic had a comparatively clean run through July. Claude Sonnet 5 launched at $2 per million input tokens and $10 per million output tokens — pricing set to rise to $3 and $15 respectively after August 31, 2026, giving enterprise buyers a clear, published cost structure to plan around, in contrast to the pricing guesswork surrounding Gemini 3.5 Pro.

Anthropic also resolved a separate disruption in the same window. Claude Fable 5, a higher-capability “Mythos-class” model, had been offline for 18 days after the U.S. government restricted access for foreign nationals. The export controls were lifted on June 30, 2026, and Fable 5 returned to global availability on July 1 across Claude.ai, the Claude Platform API, Claude Code and Claude Cowork. Anthropic offered Pro, Max, Team and select Enterprise subscribers up to 50% of their weekly usage limits on Fable 5 through July 7 as a goodwill gesture after the outage.

The net effect: Anthropic absorbed its own access disruption and still came out of July with two shipping frontier models and published pricing, while Google’s flagship remained unreachable.

OpenAI and xAI Aren’t Standing Still Either

Google is not the only lab moving cautiously with its most capable models. OpenAI’s newest tier — GPT-5.6 Sol, Terra and Luna — remained restricted to roughly 20 government-vetted partner organisations as of July 3, 2026. That is a far narrower release than a public API launch, even if it is a very different kind of restriction than Gemini 3.5 Pro’s total unavailability.

xAI, by contrast, shipped Grok 4.5 the same day GPT-5.6’s rollout began, pricing it at $2 per million tokens — positioning it as a direct, immediately available alternative for developers unwilling to wait on either Google or OpenAI’s more cautious rollouts.

Meanwhile, DeepSeek reached general availability with DeepSeek V4, offering a 1-million-token context window under an open-weights license — half of Gemini 3.5 Pro’s rumoured context window, but actually shippable today rather than a leaked spec sheet. For budget-conscious Australian developers and startups, an open-weight model with a confirmed, working 1-million-token context window available now is a materially different proposition than a rumoured 2-million-token model with no ship date.

Frontier Model Snapshot: Who’s Actually Shipping

Laid out side by side, the gap between what is shipping and what is still promised becomes clear:

Model	Developer	Status (July 25, 2026)	Price per Million Tokens (In/Out)	Access
Gemini 3.5 Pro	Google	Delayed — 67 days past promised timeline	Unconfirmed (rumoured $1.25–$15 in / $10–$60 out)	Limited Vertex AI preview only
Gemini 3.1 Pro	Google	Shipping — current flagship	Not publicly itemised in this research	General availability
Claude Sonnet 5	Anthropic	Shipping	$2 / $10 (rising to $3/$15 after Aug 31, 2026)	General availability
Claude Fable 5	Anthropic	Shipping — relaunched July 1 after 18-day pause	Not covered in this research	Claude.ai, API, Claude Code, Claude Cowork
GPT-5.6 (Sol/Terra/Luna)	OpenAI	Restricted rollout	Not disclosed	~20 government-vetted partner organisations
Grok 4.5	xAI	Shipping	$2 per million tokens	General availability
DeepSeek V4	DeepSeek	Shipping — reached GA	Not covered in this research	Open weights, 1M-token context, general availability

Read one way, the table is unremarkable: most frontier labs are managing staged, cautious rollouts of their newest models in 2026, whether for safety review, export-control compliance, or simple capacity constraints. Read another way, it shows Google alone with a flagship model that has no public access tier of any kind, more than two months after it was first announced on stage.

Market Impact: Why Investors and Enterprise Buyers Are Watching

Global AI investment has not slowed down during Google’s delay — if anything, it has accelerated. AI companies attracted $31 billion in global funding in the second quarter of 2026, up 42% from the same quarter a year earlier, according to PitchBook data cited in industry roundups. That capital is being deployed across a genuinely competitive field, not concentrated behind a single assumed winner.

For enterprise buyers, repeated slippage changes procurement behaviour in a specific, practical way: multi-year AI infrastructure decisions are increasingly hedged across more than one model provider, rather than committed to a single vendor’s roadmap. A frontier model that is two deadlines late is a harder model to build a 12-month product plan around, regardless of how strong its eventual benchmark scores turn out to be.

Alphabet has already felt one version of this reaction. Tech-insider.org’s July 1 coverage of the original slip reported a sharp drop in Alphabet’s market value alongside the delay news. Reuters’ July 21 report on the second miss did not include a specific stock reaction in the material reviewed for this article, but the underlying concern — a flagship AI product missing its own publicly implied timeline, twice — is exactly the kind of execution risk that analysts price into a stock even without a single dramatic move.

Historical Context: Google’s Pattern of AI Launch Stumbles

This is not the first time a high-profile Google AI launch has gone sideways in public. In February 2023, Google’s Bard chatbot — the company’s first major answer to ChatGPT — made a factual error in its own promotional demo video, incorrectly describing what the James Webb Space Telescope had photographed. Alphabet shares fell roughly 8% the next trading day, wiping out about $100 billion in market value in a single session, according to Reuters’ contemporaneous reporting.

That episode became a cautionary tale inside Google about rushing a flagship AI product to market before it was ready. Gemini 3.5 Pro’s delay looks, in some respects, like the opposite failure mode: rather than shipping something broken, Google appears to be over-correcting by not shipping at all, missing two self-imposed deadlines in the process.

Both outcomes carry real costs. Bard’s rushed demo damaged confidence in Google’s AI quality control overnight. Gemini 3.5 Pro’s extended limbo is a slower, quieter version of the same problem: a flagship product that developers, investors and enterprise buyers cannot yet evaluate, more than two months after Google first put it on stage.

What This Means for Australian Businesses and Developers

The delay lands at an inconvenient moment for Google in the Australian market specifically. Australian organisations are forecast to spend more than A$33.6 billion on public cloud services in 2026, up 17.9% on 2025, with infrastructure-as-a-service spending alone climbing 24.1% to A$7.1 billion, according to Gartner data reported by ComputerWeekly — spending that increasingly assumes AI workloads as a core driver, not an add-on.

Some of that spending is already flowing directly into Google’s existing Gemini products. Macquarie Bank told the Google Cloud Next ’26 conference that close to 80% of its 5,000 staff use Google’s Gemini Enterprise tool daily, seven months after adopting it, saving a reported 130,000 staff hours, according to itnews.com.au’s coverage of the conference. Google has separately claimed that roughly 75% of its cloud customers across Australia and New Zealand use AI daily in some form, per CRN Australia’s reporting from the same event.

None of that usage touches Gemini 3.5 Pro specifically, since it is not available to deploy. Australian enterprises already standardised on Google’s stack are, in effect, waiting on the same unconfirmed timeline as everyone else. Roy Morgan’s March-quarter 2026 research found 13.6 million Australians used some AI tool in an average four-week period, with ChatGPT leading at 10.5 million users ahead of Google Gemini — a competitive gap Google will not close with its consumer or developer base by leaving its flagship model unshipped through the middle of 2026.

Predictions: What Happens Next

Where this goes from here is genuinely uncertain, since Google has not committed to a third public date. Based on the pattern established over the past two months, tech-insider.org’s analysis points to five likely developments.

• A staged rollout, not a single launch. Expect Vertex AI enterprise customers and select partners to get access weeks before any broader release, mirroring how both Fable 5 and GPT-5.6 have shipped in 2026 — through gated tiers rather than a single public cutover.
• A scaled-back spec at launch. If Google ships in August, the initial version may carry a smaller context window than the rumoured 2 million tokens, with headline features filled in through later point releases — a common pattern when labs miss self-imposed deadlines and need to ship something defensible quickly.
• Deeper multi-model hedging by enterprise buyers. Expect more Australian and global procurement teams to formally split AI workloads across at least two providers, reducing any single vendor’s negotiating leverage — a trend the A$33.6 billion Australian cloud spending forecast is already pointing toward.
• Continued recruiting pressure from rivals. Anthropic and OpenAI are likely to keep targeting Google DeepMind researchers while the delay drags on; a third slip would make further high-profile departures more likely, not less.
• Sharper investor scrutiny at Alphabet’s next earnings call. If Gemini 3.5 Pro still has not shipped by Alphabet’s next quarterly results, expect direct analyst questions about the flagship model’s status, following the market reaction reported around the original June slip.

Frequently Asked Questions

Has Google officially confirmed a new release date for Gemini 3.5 Pro?

No. As of July 25, 2026, Google has said only that Gemini 3.5 Pro is being tested with partners and “coming soon,” per Reuters’ July 21 report. No specific date has been confirmed publicly.

What happened to the leaked July 17 launch date?

It passed without a launch. The July 17 date came from leaked, third-party reporting rather than an official Google announcement, and Mashable confirmed on July 24 that there was still no sign of the model as of that date.

Is Gemini 3.5 Pro available to anyone right now?

Some reports point to a limited enterprise preview through Vertex AI, but there is no public API access, consumer access, or confirmed broad partner programme as of this writing.

What specs are rumoured for Gemini 3.5 Pro, and are they confirmed?

Reported specs include a 2-million-token context window and a “Deep Think” reasoning mode, alongside pricing estimates ranging from roughly $1.25 to $15 per million input tokens depending on the source. None of these figures have been confirmed by Google.

Why did Google ship three cheaper Gemini models instead of the flagship?

Reuters reported on July 21 that Google released three lower-cost Gemini variants while giving no timing update for Gemini 3.5 Pro. Google has not explained the sequencing publicly, though shipping smaller updates while a flagship model remains in testing is a common industry pattern.

How does this compare to Claude and GPT-5.6’s availability?

Claude Sonnet 5 and the relaunched Claude Fable 5 are both in general or near-general availability with published pricing. GPT-5.6’s most capable variants are restricted to roughly 20 government-vetted partner organisations. Gemini 3.5 Pro is the only flagship among the four with no public access tier at all.

How does the delay affect Australian businesses already using Google’s AI tools?

Existing Gemini products, including Gemini Enterprise, are unaffected and continue to be used by organisations like Macquarie Bank. The delay only affects access to the new Gemini 3.5 Pro model specifically, which no Australian customer can deploy yet.

What’s the earliest Gemini 3.5 Pro could realistically launch?

There is no confirmed date. Given two consecutive missed deadlines, tech-insider.org’s analysis suggests a staged enterprise-first rollout is more likely than a single public launch date, potentially arriving through August 2026, though Google could miss that window too.

Related Coverage

For more on the AI models shaping this story, see tech-insider.org’s ongoing coverage:

• Gemini 3.5 Pro Targets July 17 as 4 Researchers Exit [2026]
• Gemini 3.5 Pro Slips to July; Google Sheds $225B [2026]
• Claude Sonnet 5 Hits 1,866 Elo at $2 a Million [2026]
• Claude Fable 5: 18 Days Offline on Export Rules [2026]
• Grok 4.5 Launches: $2/M Tokens, Same Day as GPT-5.6 [2026]
• DeepSeek V4 Hits GA: 1M Context, Old API Dies Today [2026]
• More AI & Machine Learning coverage