At Google I/O on 19 May 2026, Google introduced the Gemini 3.5 family by shipping 3.5 Flash and telling builders that 3.5 Pro was already in internal use and would roll out the following month. That was not a leak: it sat on Google's Gemini 3.5 blog, the I/O 100-things recap, and the Cloud I/O post, and Business Insider quoted Sundar Pichai asking the crowd to wait until next month. By 13 August 2026, when this cluster was flagged, June had passed with no public Pro model ID and no replacement date, so a news item alone could not tell operators whether the flagship had slipped, been cancelled, or was about to land.
The record through 23 October 2026 is now clear enough to act on. Google's 21 July model post still described 3.5 Pro as testing with partners. Alphabet's Q2 earnings call the next day kept that language while Pichai told analysts Google would need a larger Gemini 4 base model to compete at the frontier. Subsequent Gemini API release notes, the live model catalog, and Agent Platform location tables never list gemini-3.5-pro. Google instead iterated Flash models and, on 13 October, generally available Gemini 4 Argon as the flagship.
What we checked
- What status changed for Gemini 3.5 Pro, and is there confirmation from Google or a reputable independent report? confirmed. Google itself moved 3.5 Pro from a June 2026 public rollout to indefinite partner testing, and the model never appeared in public catalogs. Independent I/O coverage recorded the original delay on stage; later independent coverage reported that Google moved on to Gemini 4 Argon instead. [1][6][4][17]
- Is Gemini 3.5 Pro generally available on the Gemini API or Vertex/Agent Platform as of 23 October 2026? refuted. No gemini-3.5-pro endpoint exists in the Gemini API model list (last updated 22 October 2026), the API changelog, the deprecations table, or Agent Platform location matrices. The remaining Gemini 3 Pro-class ID is gemini-3.1-pro-preview; the current flagship ID is gemini-4-argon. [9][5][10][11]
- Did Google replace the stalled 3.5 Pro slot by shipping Gemini 4 Argon as the next public flagship? confirmed. On 30 September 2026 Google announced Gemini 4 Argon as its new frontier model, then made gemini-4-argon generally available on 13 October 2026. Google's Argon posts do not mention 3.5 Pro; independent reporting stated that Google was supposed to launch 3.5 Pro and released Argon instead. [13][14][5][17]
What we found
- On 19 May 2026 Google launched Gemini 3.5 Flash as generally available across the Gemini app, Search AI Mode, Antigravity, the Gemini API, and Gemini Enterprise, and said 3.5 Pro was already used internally with a rollout the following month. The same June timing appears in the I/O 100-things list and in Google Cloud's I/O recap, which wrote that Gemini 3.5 Pro was in testing and would be coming next month. Business Insider reported Pichai telling the I/O audience, quote, give us until next month to get it to you. [1][2][3][4]
- The June window closed without a 3.5 Pro API launch. Gemini API release notes for May through June 2026 record 3.5 Flash, Managed Agents, computer-use on 3.5 Flash, Omni Flash preview, and image-model deprecations, but no gemini-3.5-pro ID. On 21 July 2026 Google shipped 3.6 Flash, 3.5 Flash-Lite, and limited-access 3.5 Flash Cyber, and wrote that Gemini 3.5 Pro was currently testing with partners and would be made broadly available as soon as it was ready, while Gemini 4 pre-training had already started. [5][6]
- On Alphabet's 22 July 2026 Q2 earnings call, Pichai repeated that 3.5 Pro was in testing and that Google had started its most ambitious pre-training run yet for Gemini 4. In Q&A he said coding and agentic coding still needed work, that 3.6 Flash had jumped over ten points versus 3.5 Flash on a DeepSWE-style coding benchmark, and that competing at the next frontier would require Gemini 4 as a larger base model, with subsequent releases almost at a monthly cadence. Google's edited CEO remarks blog matches the prepared 3.5 Pro and Gemini 4 lines. [7][8]
- Public surfaces still do not expose 3.5 Pro. The Gemini API models page, last updated 22 October 2026, lists gemini-4-argon as the Gemini 4 flagship and, among Gemini 3 models, 3.8 Flash, 3.6 Flash, 3.5 Flash-Lite, and gemini-3.1-pro-preview, not gemini-3.5-pro. The deprecations table likewise never records a 3.5 Pro release or shutdown. Agent Platform location tables list gemini-4-argon and gemini-3.1-pro-preview, plus 3.5 Flash and 3.5 Flash-Lite, with no 3.5 Pro ID. One Cloud security guide updated 20 October 2026 still says Agent Platform includes models such as Gemini 3.5 Pro and Flash, but it gives no model code and the location matrix contradicts a live endpoint. [9][10][11][12]
- Google filled the Pro gap with Flash cadence, then deprecated the original 3.5 Flash pointer. Changelog entries show 3.6 Flash and 3.5 Flash-Lite on 21 July, 3.7 Flash on 13 August, and 3.8 Flash on 2 September 2026. On 8 October, gemini-3.5-flash was marked deprecated, with Google AI Studio selection removed and API requests automatically routed to gemini-3.6-flash. Builders who waited for a 3.5 Pro-class reasoning model were left on gemini-3.1-pro-preview plus successively newer Flash SKUs. [5][10]
- Gemini 4 Argon is the model Google actually shipped into the flagship slot. The 30 September 2026 blog announced Argon first to trusted cyber defenders in the Fairwind Program at an introductory $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. The 13 October blog and API changelog made gemini-4-argon generally available for paid API customers, Antigravity, Gemini for business, and Gemini app / Search AI Mode subscribers. The DeepMind model card describes Argon as a new frontier sparse MoE, not a modification or fine-tune of a prior model, with a unified 1M input-output window. Engadget reported that Google was supposed to launch 3.5 Pro but moved on and released Argon instead. [13][14][5][15][16][17]
What it means
Do not design, test, or document against a gemini-3.5-pro model string. Treat 3.5 Pro as an unreleased SKU. For Pro-class, long-horizon, or high-stakes reasoning work, the supported Google IDs are gemini-4-argon (stable since 13 October 2026; introductory API price $2/$10 per million input/output tokens through 30 November 2026, then $4/$20) and, if you still need a Gemini 3 Pro-class endpoint, gemini-3.1-pro-preview. For volume agent traffic, pin 3.8 Flash or 3.5 Flash-Lite rather than the deprecated gemini-3.5-flash alias, which now routes to 3.6 Flash. Ignore leftover Cloud copy that names Gemini 3.5 Pro without an ID.
The builder lesson is roadmap hygiene. Google publicly dated a Pro launch, missed it, kept a coming-soon line for two months, then shipped a new generation without a first-party cancellation post. Feature-flag the model ID, keep a fallback (3.1 Pro preview or a non-Google frontier model), and treat conference month-ahead Pro dates as intent, not a contract. Argon's 1M unified window, paid-tier API gate, and Fairwind cyber-permissive variant are the actual product changes to plan around.
Still open
- Google's Argon blogs never say 3.5 Pro is cancelled; a widely reported spokesperson remark to that effect could not be opened here because the Reuters pages returned 401.
- Whether any partner-only 3.5 Pro checkpoint still runs inside Google or Fairwind-style programs is not documented on the pages reviewed.
- The Argon model card says the model is not a fine-tune of a prior model, but Google has not published what happened to the 3.5 Pro training run.
- gemini-3.1-pro-preview still has no announced shutdown date; it is unclear how long that Gemini 3 Pro-class endpoint will remain after Argon GA.