benchr
On July 1, 2026, this page published a full review of a model called Gemini 3.5 Pro: a claimed June 30, 2026 launch, a 2,000,000-token context window, tiered pricing at $2.50/$15 per million tokens, and a set of benchmark scores including a claimed 95.5% on GPQA Diamond. A routine fact-check on July 8, 2026 found none of it holds up. Gemini 3.5 Pro has not launched.
Two checks settle it. Google's current Gemini model guide and pricing page list callable models such as Gemini 3.6 Flash and Gemini 3.5 Flash-Lite, but no Gemini 3.5 Pro endpoint or rate card. Google DeepMind's own model hub labels “3.5 Pro coming soon.” The two external sources originally cited for this page's launch and benchmark claims still do not resolve.
Claimed record versus the official record
This correction rests on records a reader can check, not on a replacement estimate. The table preserves the withdrawn claims so the error remains auditable, then shows what Google's own product and pricing pages contained when benchr checked them on July 29, 2026.
| Field | Withdrawn claim | Official record |
|---|---|---|
| Availability | Released June 30, 2026 | Not listed as a callable Gemini API model |
| API identifier | A production model existed | No provider model ID published |
| API price | $2.50 input / $15 output per million tokens | No official rate card published |
| Context | 2,000,000 tokens | No official model specification published |
| Benchmarks | 80.0 ARC-AGI-2; 95.5 GPQA Diamond | No provider result sheet for this model |
“Not listed” is a time-bounded finding, not proof that Google will never release the name. Until a callable model, model card, and rate card exist, this URL functions as a correction notice rather than a current product review.
This wasn't a real launch with an outdated number attached to it. It was a page describing a product that, as of this correction, Google has not shipped. benchr isn't going to guess at when it might arrive. Some third-party reports have floated dates, but none of them trace back to an official Google source, so none of them are printed here. If Gemini 3.5 Pro ships with verified official pricing and benchmarks, benchr will publish a new review built from that information, not a rewrite of this one.
What's withdrawn: the original review text, its benchmark table, its pricing breakdown, and the Review and Offer structured data that described a rating and a price for a product that isn't on the market. None of that belongs on a page benchr expects readers to trust.
If you came here looking for Google's actual current flagship model, that's Gemini 3.1 Pro — reviewed with real, sourced pricing and benchmarks, still current as of this correction. Every other correction benchr has made, factual or stale, is logged at /corrections.
The site's whole premise is that every figure gets checked. Publishing figures for a model that didn't exist is the exact failure that premise exists to prevent, so this correction says so plainly instead of quietly editing the page.