# Google's Gemini 3.6 Flash Cuts Coding-Agent Token Costs by Up to 65% **Google DeepMind priced its new Gemini 3.6 Flash model to cut coding-agent token costs, while a larger Gemini 3.5 Pro model stays in partner testing.** ## What Google released Google DeepMind released three new proprietary models: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber. According to [VentureBeat](https://venturebeat.com/technology/googles-gemini-3-6-flash-model-cuts-ai-agent-token-costs-by-up-to-65-on-long-horizon-engineering-tasks-and-3-5-pro-is-on-the-way), the models are built to make coding agents faster, smarter, and cheaper at scale. Gemini 3.6 Flash and Gemini 3.5 Flash-Lite are available now through the Gemini API in Google AI Studio and Android Studio, plus the consumer Gemini app and Google Search. Gemini 3.5 Flash Cyber, a specialty model for cybersecurity researchers and red teamers, has no listed price yet and will reach governments and trusted partners through Google's CodeMender bug-fixing agent, according to a separate Google blog post. ## Pricing and token savings Google prices Gemini 3.6 Flash at $1.50 for every million input tokens and $7.50 for every million output tokens, while the new Gemini 3.5 Flash-Lite costs $0.30 per million input tokens and $2.50 per million output tokens. That undercuts Gemini 3.5 Flash, priced at $1.50/$9.00 per million tokens, and Gemini 3.1 Pro Preview, priced at $2/$12. Google's older Gemini 3.1 Flash-Lite still holds the "most cost-efficient" title at $0.25/$1.50 per million tokens. On long-horizon engineering benchmarks such as DeepSWE, which grades multi-step agent task completion, the token savings from Gemini 3.6 Flash reach up to 65% versus Gemini 3.5 Flash. Across the board, the independent Artificial Analysis Index measured a 17% drop in output token usage for the new model compared with its predecessor. ## Key facts - Gemini 3.6 Flash: $1.50/$7.50 per million input/output tokens; up to 65% fewer tokens than Gemini 3.5 Flash on long-horizon engineering tasks. - Gemini 3.5 Flash-Lite: $0.30/$2.50 per million input/output tokens. - The Artificial Analysis Index measured a 17% drop in output token usage versus the prior Gemini 3.5 Flash. - Gemini 3.6 Flash and Gemini 3.5 Flash-Lite both share a one-million-token input context window, a 64,000-token max output limit, and a March 2026 knowledge cutoff. - Gemini 3.6 Flash scores 49% on DeepSWE (up from 37%) and 63.9% on MLE-Bench (up from 49.7%). - Google's own benchmark figures for these models are self-reported, not independently run. ## Benchmark gains Gemini 3.6 Flash improved on other measures Google cited too. Its OSWorld-Verified score, covering the model's built-in computer-use tool calls through the Gemini API and Gemini Enterprise, rose to 83.0% from 78.4%. On GDPval-AA v2, a knowledge-work benchmark, the model's score moved from 1349 to 1421. Google says the gains come from the model taking fewer reasoning steps and tool calls to finish multi-step workflows, plus reduced "verbosity," though it did not detail the exact technical changes behind that. Google also layered on Frontier Safety safeguards meant to harden the models against jailbreaks and misuse in chemical, biological, radiological, and nuclear domains, plus cyber offense, while training the models to avoid unnecessary refusals for legitimate requests. ## The missing flagship: Gemini 3.5 Pro Developers on social media flagged a conspicuous absence: Google's larger, more capable Gemini 3.5 Pro model, which the company had signaled would arrive this summer. Gemini 3.1 Pro, the prior flagship, debuted back in February 2026, and rivals including OpenAI and Anthropic have since shipped several newer flagship generations. Responding to the question on X, Google technical staffer Logan Kilpatrick wrote that Gemini 3.5 Pro "is currently testing with partners" and that Google plans to make it broadly available once it is ready. ## What you can check before switching If your team runs coding agents on Gemini today, compare Gemini 3.6 Flash's $1.50/$7.50 per-million-token pricing and its DeepSWE and MLE-Bench scores against whatever model you currently use, since Google reports these as self-run figures rather than independently verified results. Check the March 2026 knowledge cutoff and the one-million-token context window against your workload's needs before migrating a production agent, and watch for Gemini 3.5 Pro's general release if your use case needs a larger flagship model rather than the Flash tier. ## Check your understanding - How much cheaper is Gemini 3.6 Flash than Gemini 3.5 Flash on long-horizon engineering tasks? - What benchmark scores did Gemini 3.6 Flash post on DeepSWE and MLE-Bench compared with the previous version? - Why is Gemini 3.5 Pro still missing from Google's lineup, according to Logan Kilpatrick? ## Sources - [VentureBeat: Google's Gemini 3.6 Flash model cuts AI agent token costs by up to 65%](https://venturebeat.com/technology/googles-gemini-3-6-flash-model-cuts-ai-agent-token-costs-by-up-to-65-on-long-horizon-engineering-tasks-and-3-5-pro-is-on-the-way) --- - Category: know (pricing) - Affects: agent builders ยท tool costs - Published: 2026-07-21T19:59:58.000Z - Original: https://venturebeat.com/technology/googles-gemini-3-6-flash-model-cuts-ai-agent-token-costs-by-up-to-65-on-long-horizon-engineering-tasks-and-3-5-pro-is-on-the-way - Generated by tinker news