Last checked: August 11, 2026. Gemini 3.5 Pro is announced, but it is not publicly available. Google DeepMind’s live Gemini page still says “3.5 Pro coming soon”. Google originally expected to roll it out in June; its newer wording says the model is testing with partners and will become broadly available when ready. There is still no documented public model ID, dedicated API page, price, model card or confirmed release date.
The delay is real. The popular claim that Google has silently cancelled Gemini 3.5 Pro is not confirmed. The best evidence-based conclusion is less dramatic: Google missed its earlier expectation, the model remains in partner testing, and the public product record still points to a future release. Developers, marketers and AI buyers should not build around an undocumented endpoint; they should use Google’s shipped models today and keep migration behind a tested configuration layer.
Gemini 3.5 Pro status: the short answer
| Question | Verified answer on August 11, 2026 |
|---|---|
| Has Google announced Gemini 3.5 Pro? | Yes |
| Is it publicly available? | No documented public release was found |
| What does Google’s live model page say? | “3.5 Pro coming soon” |
| What is its current testing stage? | Partner testing, according to Google’s July update |
| Is there a public model ID? | No documented `gemini-3.5-pro` endpoint |
| Is there official pricing? | No Gemini 3.5 Pro entry |
| Is there a model card? | No dedicated card found |
| Has Google confirmed cancellation? | No |
| Has Google confirmed a new release date? | No |
The absence findings come from Google’s live Gemini family page, the Gemini API model catalog, developer pricing table, API changelog and DeepMind model-card index. They are time-sensitive: a launch can change all five pages quickly.
What Google announced—and what changed
May 19: internal use and a June expectation
Google’s May 19 Gemini 3.5 launch post focused on the publicly shipped Gemini 3.5 Flash. At the end, Google said it was hard at work on 3.5 Pro, that the model was already being used internally, and that it looked forward to rolling it out the following month.
“Next month” established a June expectation. It was not a guaranteed date, API commitment or availability promise for every region and product. June passed without a public 3.5 Pro release, which is why “Gemini 3.5 Pro delayed” is a fair description.
July 21: partner testing and no month
Google’s July 21 model update changed the status language. It said 3.5 Pro was testing with partners and would become broadly available as soon as it was ready. The month disappeared. That is the latest detailed corporate status found in this review.
The same paragraph says Google started its most ambitious pre-training run for Gemini 4. Both statements appear together: 3.5 Pro partner testing and Gemini 4 research. Gemini 4 therefore does not, by itself, prove that 3.5 Pro is cancelled or replaced.
August 11: still coming soon, still no public endpoint
At the publication check, DeepMind’s page still displayed “3.5 Pro coming soon.” The public API catalog listed Gemini 3.6 Flash, 3.5 Flash, 3.5 Flash-Lite and 3.1 Pro Preview, but not Gemini 3.5 Pro. The pricing page and model-card index likewise contained no Pro entry. That combination makes “announced but not public” the defensible status.
Do not confuse the Gemini model names
| Name | What Google documents | What it does not prove |
|---|---|---|
| Gemini 3.5 Pro | Announced flagship; partner testing; coming soon | No public ID, price, card, limits or release date |
| Gemini 3.5 Flash | Shipped model with public `gemini-3.5-flash` endpoint | Not proof that Pro shipped or was distilled into Flash |
| Gemini 3.5 Flash-Lite | Public high-volume, efficiency-oriented model | Not a Pro-class replacement claim |
| Gemini 3.6 Flash | Public newer Flash workhorse for coding, knowledge and multimodal tasks | Release sequencing does not confirm Pro cancellation |
| Gemini 3.5 Flash Cyber | Separate specialist built on 3.5 Flash with limited trusted-partner access | Not Gemini 3.5 Pro and not general API availability |
| Gemini 3.1 Pro Preview | Current documented Pro-class API model | A preview lifecycle is not a promise about 3.5 Pro |
| Gemini 3.1 Deep Think | Specialized reasoning mode built on 3.1 Pro | Not a hidden 3.5 Pro endpoint |
| Gemini 4 | Google confirms a major pre-training run | No confirmed public model or 3.5 Pro replacement decision |
This naming map prevents three common mistakes: treating a Google AI subscription called “Pro” as model access, pasting an invented `gemini-3.5-pro` string into production code, and assigning leaked benchmark screenshots to an identity the poster cannot verify.
Is Gemini 3.5 Pro cancelled?
Verdict: cancellation is unconfirmed. A SemiAnalysis claim that the model was silently cancelled was amplified on X and Reddit on August 10. That report is newsworthy because it addresses the missing launch, but Google has not published a cancellation notice. Its live model page still says coming soon, and its most recent corporate statement says partner testing continues.
There are several possible outcomes: Google could ship under the existing name, rename the checkpoint, fold work into another release, limit the rollout, or cancel it. None is confirmed. The article should therefore not promote “cancelled” from a reported claim to a product fact, nor declare the rumor permanently false. The status can change.
Why might Gemini 3.5 Pro be delayed?
Google has confirmed continued testing, not the root cause. The explanations below are competing hypotheses with different evidence strengths.
1. Coding and agentic readiness—medium confidence
Bloomberg reported, citing people familiar with the matter, that the launch was months behind schedule and coding capability was a sticking point. This is the clearest reported explanation, and it fits Google’s emphasis on coding, agents and long-horizon work across the 3.5 family. Google has not publicly named coding as the delay cause, so it remains attributed secondary reporting.
2. Partner testing and integration—medium-low confidence
Partner testing is confirmed. Such testing can expose reliability, latency, tool-use, safety and integration issues that benchmarks miss. But Google has not said partner feedback caused the delay. “Currently testing” describes a stage, not a postmortem.
3. A portfolio moving around Pro—medium-low confidence
Google shipped 3.6 Flash and 3.5 Flash-Lite while Pro remained in testing. The live family page emphasizes efficiency and production agents. That sequencing may reduce pressure to release a flagship before it meets its bar, but it does not prove resources were diverted or Pro was deprioritized.
4. Safety and evaluation readiness—low confidence
A frontier release needs safety evaluations, misuse testing, deployment controls and a model card. No 3.5 Pro card is public. That absence is normal before launch and does not establish that a safety failure caused the delay. Treat safety readiness as a possible gate, not a sourced explanation.
5. Gemini 4 changes the plan—low confidence
Google confirms Gemini 4 pre-training. Community posts infer that the next generation makes 3.5 Pro obsolete. The stronger counterevidence is Google’s own July wording, which announced both activities in parallel. A rename or replacement is possible; it is not established.
The most evidence-consistent synthesis is modest: the model did not meet Google’s broad-release bar on the original timeline; reported coding/agentic quality may be part of the reason; partner testing continues while the rest of the portfolio advances. That is an inference, not an official Google explanation.
What we can expect—and what remains unknown
Google positions “Pro” models for complex tasks, and the 3.5 family launch emphasizes agentic coding, multimodal understanding, long-horizon work and multi-step problem solving. It is reasonable to expect 3.5 Pro to target those areas. It is not reasonable to copy the specifications or benchmark scores of 3.5 Flash, 3.6 Flash or 3.1 Pro onto an unreleased model.
| Capability area | Confidence | Evidence boundary |
|---|---|---|
| Flagship complex reasoning | High as intended positioning | “Pro” and Google launch language support the target; delivered quality is unknown |
| Coding and agentic work | High as a priority | Central to the family; no official Pro score exists |
| Multimodal inputs | Medium | Strong family direction; exact Pro modalities are unpublished |
| Long-context tasks | Medium | Family capability; no Pro limit is documented |
| Tool use and managed agents | Medium | Platform direction supports it; the Pro tool matrix is absent |
| Latency, throughput and commercial terms | Unknown | No public technical or pricing page |
| Safety profile and deployment restrictions | Unknown | No dedicated model card |
Also unknown are the model’s eventual name, public endpoint, knowledge cutoff, input/output limits, regions, lifecycle status, AI Studio and Vertex availability, thinking controls, caching, batch options, quotas and subscription entitlements. Any article claiming those details today needs a source stronger than a screenshot or inferred endpoint.
What developers and agent builders should use today
Do not make an unreleased model a dependency. Choose from documented models, validate on your own task set, and keep the router replaceable.
| Current need | Documented option to evaluate | Caveat |
|---|---|---|
| Production coding, knowledge work and multimodal agents | Gemini 3.6 Flash | Validate its tool behavior, latency and quality on your harness |
| High-volume extraction, classification and bounded workers | Gemini 3.5 Flash-Lite | Use acceptance tests; low unit price does not excuse weak output |
| Existing validated 3.5 deployment | Gemini 3.5 Flash | Do not migrate only because a newer name exists |
| Current Pro-class reasoning | Gemini 3.1 Pro Preview | Preview lifecycle and terms require monitoring |
| Specialized science/engineering reasoning | Gemini 3.1 Deep Think through its documented surface | It is not a generic 3.5 Pro substitute or endpoint |
DMT’s guide to production AI-agent harness controls explains why the surrounding system—context, tools, state, retries, receipts and evaluation—often matters more than waiting for one model. A stronger unreleased model cannot fix a brittle harness today.
A release-ready migration pattern
- Put model IDs in configuration. Do not scatter them across prompts, code and automation tools.
- Pin a documented identifier. Avoid aliases whose target can change without your review.
- Build a representative evaluation set. Include normal cases, edge cases, tool failures and refusal cases.
- Measure accepted outcomes. Track factual errors, structured-output validity, tool-call accuracy, latency, usage and human rework.
- Keep a fallback. A production workflow should survive a quota issue, preview retirement or regional gap.
- Canary the real endpoint. If Google releases 3.5 Pro, route a small share of low-risk tasks before broad migration.
- Review data and lifecycle terms. Model quality is only one procurement dimension.
Compare models through a cost-per-accepted-result model evaluation, not one impressive answer. If you use parallel workers, adapt the receipt and stopping rules in DMT’s Codex subagent orchestration checklist without assuming a Google or OpenAI implementation is identical.
What the delay means for marketers
Marketing teams should not pause content research, analytics, creative prototyping or site work while waiting for 3.5 Pro. Use a documented model that passes the task, retain human review for brand and factual risk, and preserve the prompt/evaluation package so a future model can be tested quickly.
- Content and SEO: test source extraction, brief quality, claim traceability and edit time. Do not assume a flagship automatically understands your site inventory or search intent.
- Analytics: validate calculations and schema handling against known answers before allowing narrative recommendations.
- Creative workflows: compare instruction-following, brand consistency and revision rounds rather than launch benchmarks.
- Automation: separate read-only analysis from external writes, set approval gates and log every side effect.
- Buying decisions: require the vendor to name the exact model ID, lifecycle, data terms, region, quotas and commercial basis.
For the broader adoption layer, DMT’s AI-in-digital-marketing implementation guide covers real workflow design. Google’s search product is a different surface again; use the practical Google AI Mode guide for marketers rather than treating a future Gemini API model as a search-ranking feature.
What the delay means for AI buyers
“Powered by Gemini Pro” is not a sufficient procurement answer. Ask for the exact callable identifier and where it runs. Confirm whether the product uses a preview, a generally available endpoint, a managed Google surface or a vendor-controlled router. Require documented data retention, model-improvement terms, regional processing, tool access, quota behavior, incident response and deprecation policy.
Do not budget with rumored 3.5 Pro rates. Use the current official pricing for the model you can call, then run a sensitivity case for quality, latency and human review. An evidence-led frontier model comparison is helpful only when model versions, harnesses and prices are explicit.
What X and Reddit are actually saying
The bounded community sample shows more frustration than verified access. Official DeepMind and named Google builder posts moved from a June expectation to partner testing. Community posts then filled the information gap with Arena identity guesses, broken AI Studio links, benchmark screenshots, jokes and cancellation theories.
Recurring Reddit questions were practical: Is there a real endpoint? Why did the expected window pass? Will coding quality improve? What will it cost? Is Flash secretly Pro? Does Gemini 4 replace it? The healthy part of the discussion is skepticism toward unverified screenshots. The unreliable part is treating a model’s writing style, an alias or a code string as identity proof.
No community post is used here to prove a product fact. Exact links, dates, handles, subreddits and evidence classes are preserved in the package ledger. X search is personalized and incomplete; Reddit anecdotes are not a representative user study.
What to monitor next
- DeepMind model page: a change from “coming soon” to a dedicated Pro page.
- Gemini API catalog: a documented callable identifier and lifecycle.
- Pricing page: Pro-specific input, output, caching, batch and grounding terms.
- API changelog: preview or GA release entry.
- Model-card index: intended use, evaluations, limitations and safety analysis.
- Vertex AI documentation: region, quota, enterprise and managed-service details.
- Official Google/DeepMind post: a launch, rename, cancellation or revised rollout statement.
When any one of those changes, recheck all the others. A launch blog without an API entry may describe a limited product surface; an API ID without a model card or stable lifecycle may still be preview-only.
Methodology and evidence labels
This analysis distinguishes four evidence classes. Confirmed by Google means a Google/DeepMind product page, developer document, model card, changelog or named first-party announcement states the claim. Reported means a reputable publication attributes information to sources but Google has not confirmed it. Community rumor means a public post, screenshot or inference without auditable product identity. Editorial inference means a conclusion drawn from confirmed status and reported context; it is labeled as such.
The product pages, API catalog, pricing, changelog and model-card index were checked on August 11, 2026. Keyword Planner showed 1,600 average monthly India searches for `gemini 3.5 pro` and 390 for `gemini 3.5 pro release date`; DMT Search Console had no established Gemini query signal. Planning volumes are directional, not a promise of traffic.
Frequently asked questions
Is Gemini 3.5 Pro available now?
No documented public release was found on August 11, 2026. Google says the model is testing with partners, and its live page still says coming soon.
What is the Gemini 3.5 Pro release date?
Google has not supplied a current date. Its May post expected the following month, but that window passed; the July status says broad availability will come when the model is ready.
Was Gemini 3.5 Pro cancelled?
Cancellation is reported and discussed, but it is not confirmed by Google. Current first-party pages still describe the model as coming soon and in partner testing.
Is `gemini-3.5-pro` a valid public API model ID?
It was not listed in Google’s public Gemini API catalog at the publication check. Do not put an invented or leaked identifier into production code.
Is Gemini 3.5 Flash Cyber the same model?
No. Google documents Flash Cyber as a separate cybersecurity specialist built on Gemini 3.5 Flash and intended for limited trusted-partner use.
Does Gemini 4 mean 3.5 Pro will not ship?
No. Google confirmed Gemini 4 pre-training and 3.5 Pro partner testing in the same update. A future replacement is possible but not established.
What should I use while waiting?
Evaluate documented Gemini 3.6 Flash, 3.5 Flash-Lite, 3.5 Flash or 3.1 Pro Preview according to your task, lifecycle and commercial requirements. Keep model routing configurable and validate any future Pro endpoint on a fixed test set.
Author and editorial accountability
Tayeeb Khan writes Digital Marketer Tayeeb’s source-led coverage of AI, SEO and marketing workflows. This analysis uses live Google product and developer documentation, labeled secondary reporting, a bounded X/Reddit evidence sample, free keyword evidence and a fresh DMT duplicate review. It does not claim insider access to Gemini 3.5 Pro.
The practical decision: build now, keep the model replaceable
Gemini 3.5 Pro may still become an important flagship. The delay can even be positive if partner testing improves coding reliability, tool use, safety and product integration before a broad release. But optimism is not an endpoint. Teams should ship on documented models, preserve evaluation evidence, and make the eventual migration an engineering decision rather than a rumor-driven rewrite.
For now, the headline remains precise: Gemini 3.5 Pro is coming soon according to Google, but it is not public, its release date is unknown, and cancellation is unconfirmed. The next trustworthy update will arrive in Google’s model page, API catalog, pricing, changelog or model card—not in an unaudited benchmark screenshot.
