Gemini 3.5 Pro Release Date: Cancelled, Replaced by Gemini 4 Argon

Updated 2026-10-26: Gemini 3.5 Pro has been cancelled and Gemini 4 Argon has shipped in its place. We rewrote the status, added a "what to use instead" table with current pricing, and flagged the 8 October deprecations of gemini-3.5-flash and gemini-3.7-flash.

If you searched "Gemini 3.5 Pro release date," the answer has changed since our last update. Through August, this page tracked a delayed model: announced at Google I/O in May, promised for June, and repeatedly missed. On 1 October, Reuters reported that Google had stopped waiting. The company no longer plans to release Gemini 3.5 Pro, and its new top model, Gemini 4 Argon, now sits where 3.5 Pro was supposed to go.

For engineering teams this settles a planning question that had been open since May. Anything in your roadmap gated on "when 3.5 Pro ships" should now be re-pointed at a model that exists. This page is now Codersera's canonical Gemini 3.5 Pro reference. It covers what Google actually said, the full timeline, what happened to the rest of the 3.5 family, and which model to run instead, with prices checked against Google's pricing page as of October 2026.

Has Gemini 3.5 Pro been cancelled?

Yes. In a Reuters report published 1 October 2026 (syndicated by Khaleej Times), a Google spokesperson said the company no longer plans to release Gemini 3.5 Pro. The same report notes that CEO Sundar Pichai had originally said the model would be released in June.

Google's own developer surfaces line up with that:

• The Gemini API model list has no gemini-3.5-pro or gemini-3.5-pro-preview entry. The only 3.5 text model still listed is Gemini 3.5 Flash-Lite, plus the specialised 3.5 Live Translate and 3.5 Transcribe models.
• The pricing page carries no 3.5 Pro price.
• The Gemini API changelog never mentions 3.5 Pro.
• The Google DeepMind Gemini page, which carried a "3.5 Pro coming soon" marker in earlier versions of this page, has no 3.5 Pro entry. It now leads with Gemini 4 Argon.

In earlier versions of this page we drew a line between "delayed" and "cancelled" because Google was still saying 3.5 Pro was coming. That distinction no longer applies. The model is cancelled.

What replaced Gemini 3.5 Pro?

Gemini 4 Argon. The rollout came in two steps:

• 30 September (reported 1 October): According to Reuters, Google announced Argon as the top-tier model of its Gemini 4 generation. A spokesperson said it is larger than Google's previous "Pro" models and called it "our most performant model yet built for complex workloads." At that point it was going only to select cybersecurity partners, with no public release timeline given.
• 13 October: The Gemini API changelog lists "Gemini 4 Argon (gemini-4-argon) generally available (GA)," described as "our most capable model built for complex coding, agentic, math, and science tasks." The DeepMind models page also says Argon is "available today for Google AI Pro and Ultra subscribers."

Key specs from Google's Argon model page and pricing page as of October 2026:

On benchmarks: Reuters describes Google's numbers as self-reported in a press release. Per Reuters, Google described Argon as "comparable to" OpenAI's Astra and Anthropic's Opus on key coding and cyber benchmarks. On Google's self-reported numbers it beat both on several benchmarks but trailed on others, including two of the four coding benchmarks Google chose to publish. Independent evaluations were still thin when we checked, so we are not repeating any specific scores here. Run your own evals on your own workload before you migrate. For the full model breakdown, see our Gemini 4 Argon complete guide.

The full Gemini 3.5 Pro timeline

The record from announcement to cancellation:

Bloomberg's original reporting pointed to coding performance as the reason for the delay. Reuters adds broader context: during the delays Google overhauled DeepMind, and several leaders of the Gemini model team left. Over the same period Google shifted its messaging from leading capability toward cost advantage over competitors. Our older Gemini 3.5 release date and what's new post holds the original pre-launch coverage from I/O.

What happened to the rest of the Gemini 3.5 family?

Pro is not the only 3.5 model to disappear. Teams with 3.5-era IDs hardcoded should check their configs:

• gemini-3.5-flash is deprecated (8 October 2026). The changelog says it "is deprecated and has been replaced by gemini-3.6-flash." Requests are routed automatically to 3.6 Flash, and Google also suggests gemini-3.8-flash as a replacement. Your calls still succeed, but they are now served by a different model at a different price, and your quality and latency baselines no longer hold.
• gemini-3.7-flash is also deprecated (8 October 2026). It is replaced by gemini-3.8-flash, no longer selectable in AI Studio, and API requests are routed automatically. If you followed our August advice to run 3.7 Flash, you are now on 3.8 Flash whether you chose it or not.
• gemini-3.5-flash-lite is still live, listed as stable at $0.30 input / $2.50 output per 1M tokens.

Auto-routing prevents outages, but it also hides change. If your evals, cost dashboards or prompt-tuning assumed a specific model, re-baseline them now. Better still, change the ID explicitly so your logs show which model actually served each request.

What should you use instead of Gemini 3.5 Pro?

There are three realistic options. Prices below are standard paid-tier rates from Google's Gemini API pricing page as of October 2026:

Notes on the table:

• Argon's output price includes thinking tokens. Batch pricing is half the standard rate ($1.00 / $5.00 introductory, then $2.00 / $10.00). The pricing page lists no prompt-length tier for Argon.
• Argon is not on the free tier. Gemini 3.8 Flash is, with Google's note that free-tier usage may be used to improve its products.
• gemini-3.1-pro-preview is still a preview model. Its model page has no deprecation notice, but preview IDs have historically been retired once a successor stabilises. Don't treat it as a long-term home.

For a deeper look at the Flash option, see our Gemini 3.8 Flash complete guide and the 3.8 Flash vs 3.7 Flash comparison if you were auto-migrated.

Worked cost example

Take a mid-sized workload of 100M input tokens and 20M output tokens per month, with all prompts under 200K tokens, at standard (non-batch) rates:

Three takeaways:

1. Until 30 November, Argon costs less than 3.1 Pro on this mix ($400 vs $440), because its output rate is $10 against $12. That makes November a cheap window to run Argon evals.
2. From 1 December, Argon costs about 1.8× as much as 3.1 Pro on short prompts ($800 vs $440). Only move a workload permanently if Argon's quality gain justifies that difference on your own tests.
3. On long prompts the gap narrows. Above 200K tokens, 3.1 Pro charges $4.00 / $18.00 and post-intro Argon charges $4.00 / $20.00. Input is identical and output differs by about 11%.

If none of the Google options fit your budget, our roundup of the cheapest near-frontier AI models compares alternatives from other labs.

How to migrate off a 3.5-era model ID

Every Gemini model in this article uses the same google-genai SDK surface, so the migration itself is small. The work is in evaluation. A minimal pattern that keeps the model ID in config:

import os
from google import genai

# Set GEMINI_MODEL to gemini-4-argon, gemini-3.1-pro-preview or gemini-3.8-flash
MODEL_ID = os.environ.get("GEMINI_MODEL", "gemini-3.8-flash")

client = genai.Client()  # reads GEMINI_API_KEY from the environment

response = client.models.generate_content(
    model=MODEL_ID,
    contents="Summarise the failure modes in this stack trace: ...",
)
print(response.text)

A practical checklist:

1. Find hardcoded IDs. Search your codebase and infrastructure config for gemini-3.5-flash, gemini-3.7-flash and any speculative gemini-3.5-pro strings. The first two are being silently re-routed. The last one never existed as a public endpoint.
2. Re-run your eval set on the replacement model before relying on it. This matters most where requests were auto-routed.
3. Re-check caching assumptions. Argon supports implicit caching only, so pipelines built around explicit context caches need a different cost model.
4. Put the 1 December Argon price change on the calendar. Budgets built on intro pricing will double for that model.
5. Log the serving model per request so the next deprecation shows up in your dashboards instead of your invoices.

What about the promised 2M context window?

Pre-launch coverage of Gemini 3.5 Pro, including earlier versions of this page, described a 2M-token context window. That never shipped. Argon, the model that replaced it, documents a 1,048,576-token window shared between input and output. Its standout feature is output length: long decode allows up to roughly 1M output tokens, against the 65,536-token default. If you were designing around 2M tokens of input, plan for about 1M plus retrieval.

Any 3.5 Pro benchmark projections or price estimates you see elsewhere were extrapolations made before the model was cancelled. They describe a product that does not exist, so we have removed them.

FAQ

What is the Gemini 3.5 Pro release date?

There isn't one. Google announced Gemini 3.5 Pro at I/O in May 2026 for a June release, missed that target, and in a Reuters report published 1 October 2026 a Google spokesperson said the company no longer plans to release it.

Is Gemini 3.5 Pro cancelled?

Yes. Google confirmed to Reuters that it no longer plans to release the model. It has no model ID, no pricing entry and no changelog entry, and DeepMind's models page has no 3.5 Pro entry.

Gemini 3.5 Pro vs Gemini 4 Argon: what's the difference?

Gemini 3.5 Pro was never released. Gemini 4 Argon is the model that took its place at the top of Google's lineup. Google describes Argon as larger than its previous Pro models. It went GA in the Gemini API on 13 October 2026 as gemini-4-argon, with a 1,048,576-token context window, priced at $2 / $10 per 1M tokens through 30 November and $4 / $20 from 1 December 2026.

Was gemini-3.5-pro-preview ever a real model ID?

Not as a public endpoint. It followed Google's naming convention for preview Pro models, but it never appeared in the published Gemini API model list or on the pricing page. Any service claiming to offer it was mislabelling another model.

What happened to gemini-3.5-flash?

It was deprecated on 8 October 2026. Google's changelog says requests are automatically routed to gemini-3.6-flash, and it also suggests gemini-3.8-flash as a replacement. gemini-3.5-flash-lite remains available.

What should I use instead of Gemini 3.5 Pro?

Gemini 4 Argon for flagship-class work, Gemini 3.1 Pro Preview if you are already on the Pro line and not ready to move, or Gemini 3.8 Flash ($0.75 / $3.75 per 1M tokens through 31 December 2026) for cost-sensitive, high-volume workloads. Prices are as of October 2026.

Is Gemini 4 Argon available on the free tier?

No. Google's pricing page lists Argon as not available on the Gemini API free tier. In the consumer apps, DeepMind says it is available to Google AI Pro and Ultra subscribers.

Why was Gemini 3.5 Pro cancelled?

Google has not given a detailed public reason. Bloomberg reported in July that the delay was driven by coding performance, after a late-June training-data change fell short. Reuters' October report places the cancellation alongside a DeepMind overhaul, the departure of several Gemini team leaders, and the arrival of Gemini 4 Argon as the new flagship.

Sources

• Gemini API changelog (Google)
• Gemini API pricing (Google)
• Gemini API models list (Google)
• Gemini 4 Argon model page (Google)
• Gemini 3.8 Flash model page (Google)
• Gemini 3.1 Pro Preview model page (Google)
• Google DeepMind Gemini models page
• Reuters via Khaleej Times: Google announces Gemini 4 flagship AI model after months of delays (1 Oct 2026)
• 9to5Google: Gemini 3.5 Pro delays (16 Jul 2026)
• Fortune: Demis Hassabis steps down in Google DeepMind shake-up (5 Aug 2026)

If your team is migrating production workloads across Gemini, Claude or GPT models and needs extra hands, Codersera can help you hire vetted remote developers who build and evaluate LLM systems day to day.