What score will Gemini 3.5 Pro achieve on the Artificial Analysis Intelligence Index, using its maximum publicly available configuration, measured within 60 days of public release?

Gemini 3.5 Pro, announced at Google I/O in May 2026 and expected for general availability in late June or July 2026, will integrate Google's Deep Think capabilities deepmind.googlecloud.google.com. Evaluated at its maximum publicly available configuration, it is highly likely to contend for or outright take the frontier lead on the Artificial Analysis (AA) Intelligence Index.

Currently, the AA Intelligence Index (v4.1) is calibrated heavily toward agentic, scientific, and coding workloads—incorporating metrics like GDPval-AA v2 and Terminal-Bench 2.1 artificialanalysis.aiartificialanalysis.ai. Gemini 3.5 Flash serves as a reliable baseline, scoring roughly 50 in its default configuration and 55 at the "high" setting artificialanalysis.aiartificialanalysis.aiartificialanalysis.ai. The current frontier leaders sit in the high-50s to low-60s; models such as Claude Fable 5 score between 60 and 65 depending on the specific display configuration artificialanalysis.aiartificialanalysis.ai, while Claude Opus 4.8 and GPT-5.5 are clustered around 55 to 61 artificialanalysis.aiartificialanalysis.aiartificialanalysis.aibenchlm.ai.

Historically, Google's flagship Pro models launch at or near the #1 spot on this index. Because 3.5 Pro represents a substantial generational step and should reliably outperform the same-generation Flash model, its maximum configuration—bolstered by Deep Think's anticipated enhancements to scientific reasoning and agentic reliability—should push it past the 60-point threshold.

The median estimate of 62.17 reflects the expectation that Gemini 3.5 Pro will achieve a definitive step above Flash's "high" score and be highly competitive with the top reported configurations of current frontier leaders. The upside tail (P75 of 65.17, P90 of 68.33) accounts for the possibility that Deep Think drives a decisive breakthrough on this reasoning-heavy index, establishing a new state-of-the-art ceiling well into the mid-to-high 60s.

Downside risks (anchoring the P10 at 56.0 and P25 at 59.33) primarily revolve around measurement and index mechanics. AA might only measure and publish a default or non-Deep-Think score within the 60-day post-release window if API access to maximum configurations is restricted. Furthermore, an index revision could compress scores before resolution, or 3.5 Pro's primary gains could be concentrated in long-context or multimodal capabilities that are not yet heavily weighted by the current AA aggregate metric.