Gemini 3.8 Flash tops price-performance Pareto frontier

Artificial Analysis published its full review of Gemini 3.8 Flash—Google's fourth Flash model in four months—on 09-02, covering the Intelligence Index, AA-Briefcase, agent and coding benchmarks, speed, and cost. The overall verdict: intelligence is markedly improved and output is higher, though per-token pricing is up; it still lands on the cost-performance Pareto frontier and remains the cheapest option at its intelligence level.

Confirmed

• Intelligence Index: scores 59 on the high-reasoning setting, up 3 points from 3.7 Flash's 56, ranking 16th among 195 models (median 36), on par with sub-max models; Artificial Analysis also released full benchmark results across high/medium/low reasoning settings
• Agent capabilities: scores Elo 1213 on the flagship agentic knowledge-work benchmark AA-Briefcase, up 79 points from 3.7 Flash's 1134; the benchmark is based on thousands of complex inputs
• Speed and efficiency: high-reasoning setting averages about 300 tokens/sec (listed as 305 tokens/sec on the review page), completing a task in 2.5 minutes, slightly ahead of GPT-5.6 Luna (2.6 minutes)
• Output volume: averages about 48k tokens per task, 30% more than 3.7 Flash; per-task time rises from 2.2 to 2.5 minutes
• Cost: high-reasoning setting costs $0.58 per Intelligence Index task, the cheapest measured at this intelligence level, but roughly 40% pricier than 3.7 Flash, mainly due to increased output tokens

Why it matters

• The model sits on Artificial Analysis's cost-performance Pareto frontier, meaning it is currently the cheapest option at its intelligence level—directly relevant for cost-sensitive agent/knowledge-work workloads
• The 30% increase in output drives longer runtimes and higher costs, reflecting the new generation's "think longer, perform better" orientation; model selection requires weighing intelligence gains against higher unit prices
• Google has shipped four Flash models in four months, a clear acceleration in mid-tier model iteration

2026-09-02 ~ 2026-09-03 · 9 related posts

• [source] Gemini 3.8 Flash scores 59 on AA Index, cheapest model at its intelligence at $0.58 per task — ArtificialAnlys · 2026-09-02
• Gemini 3.8 Flash costs $0.58 per task, cheapest at its intelligence level — ArtificialAnlys · 2026-09-02
• Gemini 3.8 Flash outputs ~300 tokens/s, needs 2.5 min per task on high reasoning — ArtificialAnlys · 2026-09-02
• Gemini 3.8 Flash outputs 48k tokens per task on high reasoning, 30% more than predecessor — ArtificialAnlys · 2026-09-02
• Gemini 3.8 Flash scores 1213 Elo on AA-Briefcase, up 79 points over predecessor — ArtificialAnlys · 2026-09-02
• [source] Full Gemini 3.8 Flash benchmark results across high, medium and low reasoning — ArtificialAnlys · 2026-09-02
• [source] Gemini 3.8 Flash full benchmarks: #16 intelligence, fastest at 305 tokens/s — ArtificialAnlys · 2026-09-02
• Gemini 3.8 Flash Hits 59 on AA Intelligence Index but Burns More Output Tokens — haider1 · 2026-09-03
• Gemini 3.8 Flash hits 59 on AA Intelligence Index, closing on open frontier at 60 — cedric_chee · 2026-09-03