Gemini 3.5 Pro Is Delayed: No Release Date, No Reason to Wait

Gemini 3.5 Pro has no public release date. Google said it would arrive in June; on July 16, Reuters reported that the model was months behind schedule because its coding performance fell short. Do not plan a product, migration, or buying decision around it yet.

The release-date answer: there is no date

Gemini 3.5 Pro is still being tested, and Google has not replaced its missed June window with a new public date. The company's live model catalog lists Gemini 3.5 Flash and Gemini 3.1 Pro Preview, while its pricing page has no 3.5 Pro entry, API model ID, price, free-tier terms, context limit, or benchmark table.

That makes July 17 launch rumors useless for planning. A date becomes credible when Google publishes the model, access path, and commercial terms, not when an unnamed source or a countdown page repeats it.

The confirmed record is short:

Google's statement is careful. A spokesperson said the company is testing 3.5 Pro, an upgraded Flash model, and other models with partners. That confirms active development, but it does not confirm a release week, feature set, or price.

Why coding performance can hold up a flagship

Coding can hold up a flagship because the hard job is no longer generating a neat function. The hard job is staying correct while an agent reads a repository, chooses tools, edits several files, runs tests, notices a failure, and recovers without making the codebase worse.

That is agentic coding: sustained software work across many steps, tools, and decisions. A model can look excellent in a one-prompt demo and still fail when the twentieth action depends on the first nineteen being right. One weak plan or fabricated dependency can erase every speed gain that came before it.

The bar inside Google's own lineup is already awkwardly high for a new Pro model. Google positions the available Gemini 3.5 Flash as its strongest agentic and coding model and publishes these results:

Google also says 3.5 Flash produces output tokens four times faster than other frontier models. Those are vendor-published results, not independent proof, but they reveal the internal product problem: a flagship called Pro cannot merely be larger. It needs to beat the fast model on the expensive work buyers reserve for a flagship, while staying reliable enough for long agent runs.

Reuters, citing Bloomberg, reported that Google updated the training data in late June to improve coding and that the results fell short of expectations. That does not prove a full rebuild, a new architecture, or any rumored context-window size. It proves only that the reported quality bar was not met on the planned schedule.

For a mid-market CTO, this distinction matters more than the delay itself. A late model is inconvenient. A model that changes behavior across a long coding task can create review load, security risk, and rollback work that cost far more than its API bill.

What to use instead right now

Use Gemini 3.5 Flash for shipping, coding, and repeated agent loops. Use Gemini 3.1 Pro Preview when deeper reasoning is worth more latency and a higher token bill. Waiting for 3.5 Pro is the wrong default because both available choices already have public access paths and measurable costs.

The broader Gemini 3 lineup changes quickly, but the choice today is simple:

The price gap becomes visible on a normal agent workload using Google's current API rates. One million input tokens plus 200,000 output tokens costs $3.30 on 3.5 Flash at Standard rates. The same aggregate workload costs $4.40 on 3.1 Pro Preview when each prompt stays at or below 200,000 tokens. Flash is 25% cheaper in that example, before its latency advantage enters the calculation.

Google's available stack already supports substantial background work. Gemini Managed Agents can run multi-step jobs in remote environments, so a team does not need to freeze an agent pilot while it waits for a future flagship.

1. 1

   Baseline with 3.5 Flash

   Measure task success, review time, latency, and cost on the generally available model. A launch announcement cannot replace your own acceptance criteria.

2. 2

   Create one Pro escalation path

   Send only the failures caused by insufficient reasoning depth to 3.1 Pro Preview. Keep the rest on Flash so the comparison exposes whether extra spend buys a better outcome.

3. 3

   Keep the model ID configurable

   Store the model name in configuration instead of hard-coding it through the application. That turns a future 3.5 Pro evaluation into a controlled test rather than a rewrite.

4. 4

   Reopen the decision after launch

   Compare 3.5 Pro against the same task set only after Google publishes access, pricing, limits, and benchmarks. Adopt on measured lift, not on the Pro label.

The five proof gates before adopting 3.5 Pro

Gemini 3.5 Pro becomes worth evaluating when five public facts exist at the same time. A launch post alone is not enough.

1. An official API model ID. This proves developers can call the actual model through a documented interface, not a private partner preview or a third-party wrapper.
2. A complete price card. Input, output, caching, batch, grounding, and long-prompt rates determine whether a benchmark win can survive a production P&L.
3. Published operating limits. Context size, output limits, supported inputs, regions, quotas, and data-handling terms decide whether the model fits the system around it.
4. Comparable benchmarks. The useful question is not whether 3.5 Pro scores well. It is whether it beats 3.5 Flash and 3.1 Pro on the exact failure class that costs your team time.
5. A stable availability label. Preview is suitable for experiments with fallbacks. General availability is the stronger baseline for a customer-facing dependency.

These gates protect against two expensive mistakes. The first is pausing useful work for a future capability whose economics are unknown. The second is migrating on launch day, then discovering that the new flagship is slower, more expensive, or only marginally better on your workload.

Who should wait and who should move now

Most teams should move now, but they should move on a model-agnostic architecture.

Coding buyers should also compare outside the Gemini family. The current coding-model ranking is the better frame when repository work matters more than staying inside Google's stack.

The only defensible reason to wait is a narrow one: your project has a documented requirement that Flash and 3.1 Pro cannot meet, and delaying the project costs less than choosing another provider. Even then, treat 3.5 Pro as one future candidate, not the assumed answer.

Is Gemini 3.5 Pro delayed?

Yes. Google said on May 19 that it expected to roll out 3.5 Pro the following month. Reuters reported on July 16 that the model was months behind schedule, with coding performance short of internal expectations.

Is Gemini 3.5 out now?

Gemini 3.5 Flash is generally available through the Gemini API and Google's products. Gemini 3.5 Pro is not publicly listed in Google's model catalog or pricing page.

Is Gemini 3.5 Flash better than 3.1 Pro?

Flash is the stronger default for speed, cost, coding, and repeated agent loops. Google publishes a 76.2% Terminal-Bench 2.1 result for 3.5 Flash, while its 3.1 Pro benchmark table reports 68.5% on Terminal-Bench 2.0, so that is not a clean same-version benchmark comparison. Use 3.1 Pro Preview when deeper reasoning improves your actual task, not because Pro sounds stronger.

Is Gemini 3.5 Pro free?

No public price or free-tier term exists for 3.5 Pro. Claims of free access are not Google pricing. Gemini 3.5 Flash currently has a free API tier; Gemini 3.1 Pro Preview does not.

Get the next model decision memo when the facts change.