# Gemini 3.8 Flash Is On Pareto Frontier Of Intelligence And Cost On Artificial Analysis Intelligence Index > Source: > Published: 2026-09-02 16:09:40+00:00 Google might not be at the AI frontier, but it’s made its way back to the Pareto frontier. Google has released [Gemini 3.8 Flash](https://officechai.com/ai/gemini-3-8-flash-scores-59-on-artificial-analysis-intelligence-index-jumps-3-points-over-gemini-3-7-flash/), its fourth Flash model in under four months, and Artificial Analysis says it’s the one that finally gets the price-to-intelligence trade-off right. With high reasoning turned on, the model scores 59 on the Artificial Analysis Intelligence Index, up three points from [Gemini 3.7 Flash](https://officechai.com/ai/gemini-3-7-flash-benchmarks/), and it now sits directly on the Pareto frontier of the Intelligence vs. Cost per Task chart — meaning no other model currently on the market offers more intelligence for less money. A score of 59 puts Gemini 3.8 Flash (high) on par with the sub-maximum reasoning efforts of two much bigger rivals: GPT-5.6 Sol running at xhigh, and Grok 4.6 at medium. That’s a notable position for a model still carrying Google’s “Flash” branding, historically reserved for the cheaper, faster tier sitting below its Pro-series releases. ## Pricing Holds, Cost Per Task Climbs Google is keeping Gemini 3.8 Flash’s per-token pricing identical to its predecessor’s discounted rate through the end of the year — $0.75 per million input tokens and $3.75 per million output tokens, reverting to $1.50/$7.50 at standard pricing once the discount period ends. Cached input tokens keep the same 90% discount. Despite unchanged per-token pricing, Gemini 3.8 Flash actually costs more to run per task. Artificial Analysis puts its Cost per Task at $0.58 with high reasoning, roughly 40% higher than [Gemini 3.7 Flash’s $0.40](https://officechai.com/ai/gemini-3-7-flash-is-at-pareto-frontier-of-intelligence-vs-speed-says-artificial-analysis/). The jump comes down to the model simply doing more: output tokens per task rose 30% to an average of 48,000, and it takes more turns on agentic evaluations. Even so, $0.58 per task is still the cheapest cost Artificial Analysis has measured at this level of intelligence, comparable to GPT-5.6 Terra (max) at $0.53. Cost per Task drops to $0.41 on medium reasoning and $0.24 on low reasoning, with the low-reasoning configuration matching the intelligence score of Gemini 3.6 Flash (high, 52) at roughly 30% lower cost and about a third of the time. ## Where The Gains Are Coming From The three-point jump on the Intelligence Index is driven almost entirely by agentic performance rather than raw knowledge or reasoning gains. Artificial Analysis points to improvements across τ³-Banking (tool use), Terminal-Bench v2.1 (coding), and GDPval-AA v2 (real-world economically valuable tasks) as the main contributors. The single largest gain is on τ³-Banking, where Gemini 3.8 Flash picks up 12 points over Gemini 3.7 Flash to score 45%. On Artificial Analysis’s own AA-Briefcase evaluation — a private benchmark of realistic agentic knowledge-work tasks scored on correctness, analytical quality and presentation — Gemini 3.8 Flash posts an Elo of 1213, up 79 points from Gemini 3.7 Flash’s 1134. Most of that improvement comes from stronger rubric scoring and Analytical Quality rather than any change in how the output is presented. ## Speed Trade-Off The extra reasoning and longer agentic turns come at a modest cost in latency. At high reasoning, Gemini 3.8 Flash still outputs roughly 300 tokens per second, but Time per Task rises from 2.2 minutes (Gemini 3.7 Flash) to 2.5 minutes — just ahead of GPT-5.6 Luna (max, 2.6 minutes) but behind Claude Fable 5.1 (medium, 2.1 minutes). Drop to low reasoning, however, and Time per Task falls to just 0.8 minutes, enough to put Gemini 3.8 Flash on the Pareto frontier of Intelligence vs. Time per Task as well as cost. ## Model Details Gemini 3.8 Flash carries a 1 million token context window, unchanged from Gemini 3.7 Flash. It supports text, image, video, and speech input, though output remains text only. Across its three reasoning levels, it scores 59 on high, 57 on medium (matching GPT-5.6 Terra max and Muse Spark 1.2 xhigh), and 52 on low. With three Flash releases in the last four months — [Gemini 3.6 Flash](https://officechai.com/ai/gemini-3-6-flash-scores-50-on-artificial-analysis-intelligence-index-same-as-gemini-3-5-flash/), Gemini 3.7 Flash, and now 3.8 — Google is iterating on its budget tier faster than any other lab is iterating on comparable models. The pattern so far has been consistent: modest but real intelligence gains each cycle, paid for with a corresponding rise in token usage and Cost per Task, but never enough to knock the model off the frontier of what money can currently buy in exchange for intelligence.