Cheaper AI models that finally care about performance per dollar
Gemini 3.6 Flash and Gemini 3.5 Flash Lite are cheaper AI models in Google’s Gemini family that focus on higher coding performance, reduced token usage, and faster throughput so developers and enterprises can run real workloads at lower overall cost per task, not just chase benchmark scores. Alphabet has released three new Gemini Flash variants—3.6 Flash, 3.5 Flash Lite, and 3.5 Flash Cyber—aimed squarely at teams building AI agents that need efficiency and reliability in production. Gemini 3.6 Flash and 3.5 Flash Lite are already live in AI Studio and Vertex AI, while 3.5 Flash Cyber enters a limited pilot for governments and trusted partners. This is not a side-grade; it is Google’s clearest statement yet that the AI model marketplace is now about cost-efficiency as much as raw power.
Gemini 3.6 Flash: token-efficient coding workhorse at a lower price
Gemini 3.6 Flash is engineered to win the token cost comparison, especially for coding and data-heavy workflows. It is priced at USD 1.50 (approx. RM6.90) per million input tokens and USD 7.50 (approx. RM34.50) per million output tokens, undercutting the earlier Gemini 3.5 Flash, which sits at USD 9 (approx. RM41.40) per million output tokens. According to product leads, Gemini 3.6 Flash uses up to 17% fewer tokens than 3.5 Flash while delivering better performance per task, with a March 2026 knowledge cutoff across all context lengths. In practice, that means finance teams can process documents faster, retailers can run catalog and support pipelines more cheaply, and healthcare researchers can sift large datasets without torching their budget. On DeepSWE and MLE Bench, it makes fewer mistakes in code edits and performs better on machine learning research tasks, signaling real gains in coding performance rather than marketing gloss.
| Spec | Gemini 3.5 Flash | Gemini 3.6 Flash |
|---|---|---|
| Output token price | USD 9 (approx. RM41.40) / million | USD 7.50 (approx. RM34.50) / million |
| Input token price | Not specified in sources | USD 1.50 (approx. RM6.90) / million |
| Token usage vs 3.5 Flash | Baseline | 17% fewer tokens |
| Focus | General Flash tier | Coding, multimodal, knowledge work with better precision |
Gemini 3.5 Flash Lite: throughput king for agents and batch workloads
If Gemini 3.6 Flash is the efficient brain, Gemini 3.5 Flash Lite is the cheap, fast executor that keeps AI agents and batch jobs humming. It is priced at USD 0.30 (approx. RM1.38) per million input tokens and USD 2.50 (approx. RM11.50) per million output tokens, and is described as Google’s fastest, most cost-effective 3.5-series model for high-throughput execution. More importantly, Gemini 3.5 Flash Lite reaches 350 tokens per second, the highest output rate in the Flash family, and is explicitly tuned for high-volume, cost-conscious workflows. That makes it a natural fit for translation, content moderation, and the countless small reasoning hops inside agent systems—the places where developers feel every cent of token spend. It also outperforms previous Flash-Lite models on coding and agentic tasks, meaning you are not trading away competence for speed. For many teams, this is the model that keeps AI affordable enough to deploy widely rather than as a boutique feature.
Gemini 3.5 Flash Cyber: narrow, high-stakes value in security
Gemini 3.5 Flash Cyber is the outlier: a specialist model built for vulnerability detection and patching rather than broad developer use. It is fine-tuned to find bugs faster than human teams can, and its rollout is intentionally restricted to governments and trusted partners via CodeMender in a limited pilot to reduce misuse risks. This move signals two things. First, Google accepts that AI is now capable of discovering security flaws at a pace that demands new guardrails. Second, it believes cost-efficient Flash-tier models are strong enough to matter in cybersecurity, a space where Anthropic’s Mythos and other competitors have tried to set the tone. While ordinary developers will not touch Flash Cyber yet, its existence underscores the broader strategy: use cheaper AI models not only for routine coding performance, but also for high-stakes domains where speed and precision directly affect risk.
Why these Flash models matter now—and what comes next
Google is shipping these cheaper AI models at an awkward moment: its much-awaited Gemini 3.5 Pro is delayed, with underwhelming internal coding benchmarks forcing retraining and pushing the public launch back. Leaning on Flash while Pro “keeps cooking” is a pattern we have already seen, and 3.6 Flash is clearly built to repeat the play—keeping Google in the conversation without needing to beat every top-tier GPT or Claude variant. Meanwhile, the lab has recently slipped out of the top five on a major intelligence index as rivals flood the market with high-scoring models. In response, it is iterating rapidly: Gemini 3.5 Pro is in partner testing and should be available more widely soon, and Gemini 4 is already in pre-training. The real takeaway for developers is blunt: in a crowded marketplace, Google’s most compelling pitch is no longer "smartest model" but "best performance per dollar"—and with Gemini 3.6 Flash and 3.5 Flash Lite, that pitch is finally credible.
