Gemini 3.5 Flash is fast and much stronger on agentic tasks, but costs over 5x more to run than Gemini 3 Flash, which changes the cost calculus for Flash-tier workloads.
Key takeaways · AI-distilled
Per-tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition → prices tripled to $1.50/$9.00 per 1M input/output tokens; the rest of the cost jump came from more input tokens as agentic evalA repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.Full definition → ran more turns, while output tokens stayed flat (73M vs 72M).
On GDPval-AA, Gemini 3.5 Flash reached an Elo of 1656, ahead of Gemini 3 Flash (1204) and Gemini 3.1 Pro (1314) and just behind GPT-5.4 (xhigh, 1674), in an area Artificial Analysis calls a historical Gemini weakness.
Its AA-Omniscience score rose 11 points, mainly from fewer hallucinations, with the hallucinationWhen a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.Full definition → rate falling 31 points to 61%.
It accepts image, video and speech input with text output, and scored 84% on MMMU-Pro, the highest multimodalA model that works with more than text — reading images, audio, or video, and sometimes generating them too.Full definition → score Artificial Analysis has recorded, ahead of Gemini 3.1 Pro at 82%.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
