Google Launches Gemini 3.6 Flash and 3.5 Flash-Lite; Pro Remains in Partner Testing
Google introduced Gemini 3.6 Flash at $1.50/$7.50 and 3.5 Flash-Lite at $0.30/$2.50 per million tokens. It planned restricted Flash Cyber access, while 3.5 Pro remained in partner testing. Performance and efficiency comparisons are attributed to Google.
Enterprise agent spend is increasingly Flash-class, not flagship-only. A cheaper, less verbose 3.6 Flash plus a 350 tok/s Lite changes unit economics—while Pro partner-testing phase + Cyber lock show Google still managing capability and dual-use risk carefully. Read this as a production stack refresh, not the Pro partner-testing story ending.
Google introduced Gemini 3.6 Flash, 3.5 Flash-Lite, and a limited 3.5 Flash Cyber pairing with CodeMender. 3.6 Flash is the new workhorse for coding, knowledge work, and multimodal agents at $1.50 / $7.50 per 1M input/output tokens. 3.5 Flash-Lite targets high-throughput agentic tasks at $0.30 / $2.50 and ~350 output tokens/s (Artificial Analysis). 3.5 Pro is still partner testing only; Google also says Gemini 4 pre-training has started.
What shipped
From Google’s Keyword / DeepMind primary (Tulsee Doshi) and DeepMind Flash Cyber post:
| Model / product | Role | Price (per 1M tok) | Access day one |
|---|---|---|---|
| 3.6 Flash | Workhorse: coding, knowledge work, multimodal agents | $1.50 in / $7.50 out | Gemini API (AI Studio, Android Studio), Antigravity, Gemini Enterprise, Gemini app |
| 3.5 Flash-Lite | Fastest 3.5-class; high volume / low latency | $0.30 in / $2.50 out | Same developer surfaces; also rolling into Google Search |
| 3.5 Flash Cyber + CodeMender | Cyber find/fix at scale | Lower token price than larger cyber models (not full public list) | Governments + trusted partners only, planned soon as a limited pilot |
| 3.5 Pro | Flagship | — | Still not GA — partner testing; “as soon as ready” |
| Gemini 4 | Next gen | — | Most ambitious pre-training run yet started (no ship date) |
Vendor-claimed gains for 3.6 Flash vs 3.5 Flash include better DeepSWE (49% vs 37%), MLE Bench (63.9% vs 49.7%), OSWorld-Verified (83.0% vs 78.4%), and GDPval-AA v2 (1421 vs 1349), with computer use as a built-in client-side tool. Google says 3.6 Flash is more token-efficient and cheaper per agentic task than 3.5 Flash despite higher capability. 3.5 Flash-Lite is pitched as beating 3 Flash on several agentic/coding boards (e.g. SWE-Bench Pro 54.2% vs 49.6%; OSWorld-Verified 74.0% vs 65.1%).
Flash Cyber is explicitly dual-use controlled: multi-agent CodeMender stacks for vulnerability discovery and patching, competitive on CyberGym in Google’s charts, not open Chat/API GA.
Claims vs checks
The prices, benchmark comparisons, 17% token-use reduction and quoted 350-token-per-second throughput above come from Google’s launch account. Its AA references do not turn the entire vendor table into an independent audit. A dated independent launch ranking is not established here.
Limits
- Lab DeepSWE/MLE/OSWorld gains are vendor tables.
- Throughput and token-use comparisons depend on the test configuration and workload.
- 3.5 Pro remained in partner testing; the announcement gave no Gemini 4 release date.
Correction (October 14, 2026): Distinguished the planned Cyber pilot from immediate access and removed unsupported historical leaderboard comparisons.