Model components / Fine-tuned / chat models

Gemini 3.5 Flash

Released at Google I/O 2026 (May 19, 2026) as the new default in the Gemini app. Marketed as 'Pro-grade reasoning at Flash-level speed'; #9 on the May 2026 LMArena text leaderboard (Elo 1480) and #2 on LiveCodeBench at 90.8%. Configurable thinking budgets let users trade reasoning depth for latency. Powers Antigravity / Gemini CLI and most Google-side coding-assistant deployments where cost matters.

Google Gemini 3.5 Flash, launched at I/O 2026 on May 19, 2026 as the first Gemini 3.5 model; price-performance thinking/agentic instruct model. Available via Gemini app, AI Mode in Search, Google Antigravity, Gemini API and enterprise. Proprietary API-only. (Matches the flagship anchor 'Gemini 3.5 Flash'; scored here for the finetuned_chat category, calibrated to that anchor.) Verified live June 2026. Consolidated on 2026-07-29 from gemini-2-5-flash; openness follows gemini-3-5-flash, the release that currently governs. See sources/slug_aliases.yaml.

Openness

1 high confidence

1.0

weights: closed
data: closed
code: closed
license: Proprietary(Gemini app/Search/API/enterprise, no downloadable weights)

Proprietary and API-only across the Flash line, no downloadable weights, so 1/closed. Gemini 3.5 Flash governs as the current release and is a genuine capability jump - it scores 5 where 2.5 Flash scored 3 - which is why the versions were worth combining rather than keeping the older figure. Held apart from Gemini Pro, which Google prices and positions as a separate tier.

https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/ recorded 2026-06-04
proprietary launch across Gemini app/Search/API/enterprise, no open weights
https://deepmind.google/models/gemini/ recorded 2026-06-04
DeepMind model listing for the Gemini family; served via API and apps only, no weight downloads

Adoption

5 medium confidence

5.0

High-volume cheap tier powering the Gemini app and AI Mode in Search plus heavy API/Vertex traffic; Gemini app >750M MAU (TechCrunch, Feb 2026). Flash is the default high-volume serving model so its real usage is firmly >10M-equivalent. Attributed to the app/Search surface, not a standalone download count.

https://techcrunch.com/2026/02/04/googles-gemini-app-has-surpassed-750m-monthly-active-users/ recorded 2026-06-04
Gemini app >750M MAU
https://ai.google.dev/gemini-api/docs/models recorded 2026-06-04
Gemini 2.5 Flash positioned as high-volume low-latency production model

Capability

5 high confidence

5.0

Frontier-tier; top-right of Artificial Analysis Intelligence Index. Scored 5, consistent with the flagship anchor (cap 5). Google-reported figures.

https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/ recorded 2026-06-04
Terminal-Bench 2.1 76.2, MCP Atlas 83.6, CharXiv 84.2, GDPval-AA 1656 Elo

Unchanged since 2026-07-29 (last edited, not re-checked)