Google Just Dropped a Frontier-Class Model at Budget Pricing
On Wednesday, Google introduced Gemini 3.8 Flash, its best coding and reasoning model yet, alongside a government-tuned security twin called Gemini 3.8 Flash Cyber. The release is the third Flash model in six weeks, a rhythm that marks a clear shift inside DeepMind: after months of delays on Gemini 3.5 Pro and benchmark misses against Chinese open-weight models, Google is betting on speed, price, and agent capability to claw back its spot in the frontier race.
The headline number comes from Artificial Analysis: set to high reasoning, Gemini 3.8 Flash scores 59 on the Intelligence Index, three points above its predecessor and level with GPT-5.6 Sol and Grok 4.6. At roughly $0.58 per Intelligence Index task, it is the cheapest model at its intelligence tier, about six times cheaper per task than Anthropic's Fable 5.1 flagship.
The trade-off: it works harder, and costs a bit more per task
Google's team is candid about the design. "3.8 Flash works harder," said Tulsee Doshi and Raluca Ada Popa in the launch blog post, "executing extra reasoning steps and calling tools iteratively." That diligence is why the model is roughly 40 percent more expensive per task than its predecessor, even at a similar base price. It simply generates more tokens and takes more turns when acting as an agent. Developers who want leaner inference can dial down effort levels or stick with 3.7 Flash, which remains supported for efficiency-first workloads.
The gains show up in benchmarks. On DeepSWE v1.1, a long-horizon software engineering test, 3.8 Flash solves complex end-to-end engineering problems better than most larger frontier models at a fraction of the cost, and it scores 54.9 percent on HLE-Verified. Google also points to strength in enterprise knowledge work, including Vals Finance Agent V2, Harvey's Legal Agent Benchmark, and strong prompt-injection robustness measured by Gray Swan.
Pricing is the hook, but it is also time-limited. Introductory rates of $0.75 per million input tokens and $3.75 per million output tokens expire December 31. From January 1, 2027, those numbers double to $1.50 and $7.50.
Gemini 3.8 Flash Cyber: Security for trusted defenders only
Alongside the general model, Google released Flash Cyber, a variant tuned for finding and fixing software vulnerabilities, delivered through the new Fairwind Program to governments, critical infrastructure operators, and software maintainers. It prioritizes defense over offense. Training focused on vulnerability remediation rather than exploitation, and the Cyber version keeps a more permissive set of mitigations specifically because its users face real threats; Google's Frontier Safety Framework limits the general model's CBRN and cyber-offense use cases.
The security model's numbers are strong. It surpasses 3.5 Flash Cyber and much larger frontier models in autonomous vulnerability discovery on the CyberGym benchmark, and on an internal test spanning 20 programming languages it reaches a success rate above 70 percent. On CWE-Bench, an external patching benchmark from Collinear, it hits a pass rate of 47.2 percent versus 47.8 percent for the leading frontier model, at significantly lower cost. Google says it already uses Flash Cyber internally to secure its own code.
- Reasoning: 59 on the Artificial Analysis Intelligence Index, level with GPT-5.6 Sol and Grok 4.6
- Price per task: about $0.58, the cheapest at its intelligence tier
- Intro pricing: $0.75 per 1M input tokens, $3.75 per 1M output, until Dec 31
- Availability: Gemini API, AI Studio, Google Antigravity, Android Studio, Stitch, Gemini Enterprise, and consumer Gemini subscriptions
Adoption is already broad. Consumers on Google AI Pro and Ultra plans can reach the model through the Gemini app, AI Mode in Google Search, and Gemini in Google Sheets, which widens the reach well beyond developers into everyday use.
For enterprises, the model is a direct bid for agentic workloads, where the combination of frontier-level reasoning, tool calling, and budget pricing is the most valuable combination on the market right now.
For the moment, Google's come-up looks more credible than it did in August, when a brief slide in the leadership picture and missed flagship promises left room for doubt. Rounding out the launch, the timing also helps the narrative: the same week Google fended off the forced AdX breakup, adding to a streak of wins as it tries to turn a losing year around.
Comments