Headline: Google ships three new Gemini “Flash” models — fast and cheap, but the promised Pro edition is nowhere to be found
Google today released three new Gemini models aimed at speed and scale — Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber — but the higher‑power Gemini Pro variant that many expected after Google I/O is still missing.
What happened to Gemini Pro?
- Google had teased a Gemini 3.5 Pro at I/O in May and suggested a Pro rollout within a month. That deadline quietly slipped after internal testing found the Pro model underperforming on coding tasks, Bloomberg reports. A late‑June attempt to patch the problem by updating training data failed to bring it up to internal targets.
- The market reacted: Alphabet shares tumbled about 4.4% on the report, wiping roughly $200 billion from its market cap in one session.
- Google says Gemini 3.5 Pro will ship “as soon as it’s ready.” The last Pro-tier release was Gemini 3.1 Pro in February.
What Google did ship
1) Gemini 3.6 Flash — the headline model
- Positioning: speed-optimized, cost-effective model for AI agents and high-throughput applications.
- Efficiency: uses 17% fewer output tokens than 3.5 Flash (Artificial Analysis Index).
- Pricing: $1.50 per million input tokens; $7.50 per million output tokens (down from $9 on 3.5 Flash output).
- Benchmarks: big wins in some long-horizon engineering tests — DeepSWE v1.1: 49% (vs. 37% for 3.5 Flash); MLE‑Bench: 63.9% (vs. 49.7%). It also led OSWorld‑Verified at 83.0%, slightly ahead of Anthropic’s Claude Sonnet 5 (81.2%) and OpenAI’s GPT‑5.6 Luna (72.6).
- Where it trails: rivals remain stronger on other agent and terminal coding benchmarks. GPT‑5.6 Luna, for example, scores 67% on DeepSWE and 84.7% on Terminal‑Bench 2.1. On GDPval‑AA v2 (an Elo‑style knowledge work ranking) Claude Sonnet 5 sits at 1607 vs. 3.6 Flash’s 1421.
Hands‑on note (coding)
- In our quick coding test, Gemini 3.6 Flash produced a malformed HTML file that didn’t render properly. A third‑party tool, Deepseek, analyzed the output, found 11 bugs and applied 8 fixes to make the project playable. That suggests Gemini’s high‑level reasoning was on target, but the details and execution still need work — you can expect lots of iteration or manual fixes if you use the model for production coding today.
2) Gemini 3.5 Flash‑Lite — built for volume
- Focus: raw throughput. Claimed throughput of 350 output tokens/sec.
- Pricing: $0.30 per million input tokens; $2.50 per million output tokens.
- Use cases: document processing at scale, agentic search systems, session compaction for long agent workflows.
- Performance: despite being cheaper, it beats the older 3 Flash on several coding benchmarks, including Terminal‑Bench 2.1 (54% vs. 31%).
3) Gemini 3.5 Flash Cyber — restricted security tooling
- Availability: not public. Google is limiting access to governments and vetted partners.
- Purpose: targeted at finding and fixing software vulnerabilities — a dual‑use capability the company isn’t comfortable releasing broadly.
Where this matters for crypto
- Faster, cheaper Flash models are attractive for blockchain use cases that need scale: on‑chain data processing, large‑scale document or log analysis, high‑throughput indexers, or agentic systems that automate monitoring and alerts.
- The Cyber model’s vulnerability-finding capability is directly relevant to smart‑contract auditing and security teams — but Google’s restricted access means crypto projects and security firms may still need to rely on third‑party tools or vetted partnerships for this level of automated vulnerability scanning.
- The Pro delay is notable for projects that want heavy reasoning and reliable code generation for smart contracts or automated trading strategies — Flash series favors speed and cost over the deep, compute‑intensive reasoning that Pro models provide.
Looking ahead: Gemini 4 is already cooking
- Google confirmed it has begun an ambitious pre‑training run for Gemini 4. That means the company is actively building the next big model cycle even as it irons out issues with Pro-tier performance.
Availability
- Gemini 3.6 Flash and 3.5 Flash‑Lite are live now in the Gemini app, Google AI Studio, and via the API. Gemini 3.5 Pro remains delayed until it meets Google’s standards.
Bottom line
Google doubled down on fast, cheap models aimed at agents and high-volume workloads — a win for teams that need throughput and lower token costs. But the absence of a Pro release and lingering quality issues on coding tasks mean companies that need high‑accuracy reasoning (including many crypto security and smart‑contract workflows) may still hold off until a more capable Pro model or Gemini 4 arrives.
Read more AI-generated news on: undefined/news
