Google released Gemini 3.8 Flash on September 2, priced at $0.75 per million input tokens and $3.75 per million output through December 31. On January 1, 2027, those numbers double to $1.50 and $7.50. It's the third Flash model Google has shipped in six weeks, following the 3.6 Flash and security-gated 3.5 Flash Cyber release in late July.

The benchmark story is unusually clean for a mid-cycle Flash bump. On Terminal-Bench 2.1, 3.8 Flash scores 90.8%, up from 81.6% for 3.7 Flash. It posts 54.9% on HLE-Verified. On Artificial Analysis's Intelligence Index it registers 59 at the high-reasoning setting, at $0.58 per task, compared to roughly $3.76 for Claude Fable 5.1. The gap is the story: near-frontier output at Flash economics, at least until New Year's Day.

Google DeepMind's Tulsee Doshi and Raluca Ada Popa write that the model "delivers substantial gains from 3.7 Flash, often approaching the performance of higher-cost frontier models." They also flag the cost of that lift: "3.8 Flash works harder. On complex tasks, it exhibits greater diligence — executing extra reasoning steps, and calling tools iteratively." Artificial Analysis's numbers confirm it. 3.8 Flash burns roughly 40% more per Intelligence Index task than 3.7 Flash, which Google says "remains fully supported for efficiency-first workloads."

That's the buried line item. Sticker price is halved through year-end, but token consumption per task climbs. For agentic workloads calling tools iteratively across long horizons, the arithmetic tightens fast, and January's doubling then lands on an already inflated per-task base.

The competitive framing is Google's. Rolling out via AI Studio, the Gemini API, Antigravity, and Gemini Enterprise, 3.8 Flash is positioned against GPT-5.6 Sol and Grok 4.6 in the top tier, and against Claude Fable 5.1 on price. The Fairwind Program gates 3.8 Flash Cyber, the security-hardened variant, to approved partners.

Google Cloud CEO Thomas Kurian told CNBC that nearly three-quarters of Google Cloud customers already use its AI products and are spending roughly 50% more than their original commitments. Doshi said the recent Flash models "really surprised us in positive ways in their performance." The Street is less lyrical. D.A. Davidson's Gil Luria: "From a product perspective this model seems to keep Google in the race, but probably won't change the fact they are a distant third in the enterprise market."

The release fits a familiar pattern. When an incumbent can't win outright on capability, it competes on the cost curve, betting workloads get locked in before pricing normalizes. It's the same dynamic pulling developers toward open weights like Alibaba's Qwen3-8 27B, different substrate, same gravitational pull. The four-month window is the offer. Whether it converts is a January question.

Sources

https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/
https://www.cnbc.com/2026/09/02/google-starts-september-with-ai-momentum-after-long-losing-streak.html
https://www.theregister.com/ai-and-ml/2026/09/02/with-gemini-38-flash-google-reminds-everyone-its-still-in-the-race/5294049
https://www.datacamp.com/blog/gemini-3-8-flash-cyber
https://9to5google.com/2026/09/02/gemini-3-8-flash-launch/