Google ships Gemini 3.6 Flash, 3.5 Flash-Lite and a limited Cyber model as Gemini 3.5 Pro slips again
量子位 · wechat · 2026-07-22
Google has released three new Gemini variants—Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and a limited Cyber model for vulnerability finding and patching—while its long-delayed Gemini 3.5 Pro is still missing.
- Gemini 3.6 Flash: stronger code, AI research, and multimodal performance; DeepSWE reportedly rises from 37% to 49%, and OSWorld from 78.4% to 83.0%. Google says output token usage is down 17% vs. 3.5 Flash, and up to 65% in some coding scenarios. Pricing is $1.5/M input and $7.5/M output.
- Gemini 3.5 Flash-Lite: the fastest and cheapest Flash-Lite yet, with Google citing 350 tokens/sec output. Pricing is $0.3/M input and $2.5/M output; in some coding and agent tasks it reportedly beats the full 3 Flash model.
- Cyber: a security-focused fine-tune for government and trusted partners only.
The post also highlights ongoing frustration around Gemini 3.5 Pro, which was promised months ago but remains in partner testing, while Google says Gemini 4 pretraining has already begun and is its most ambitious training run yet.
Related event: Google Releases Three New Gemini Models, Announces Gemini 4 Pre-training(153 posts)→
More from Models
- Sakana AI Announces Project Fugu: Exploring Unified Model Coordination — SakanaAILabs · 2026-07-23
- One user says GPT-5.6 Sol Medium is now their most-used model — MatthewBerman · 2026-07-23
- Gradium boosts speech-to-text accuracy on rare names with keyword prompting — mattturck · 2026-07-23
- OpenAI's GPT-5.6 Escapes Sandbox to Hack Hugging Face During Benchmark Test — Ars Technica AI · 2026-07-23
- Fingerprint analysis says Kimi K3 and Fable 5 write more alike than sibling models — alex_verem · 2026-07-23
- Cisco says two small open security models beat GPT-5.5 on vulnerability detection cost — The Decoder · 2026-07-23