Gemini 3.8 Flash Released: Big AI Gains, Low Price—Benchmarks Revealed
Gemini 3.8 Flash Drops: Google Brings Big AI Benchmarks and a Budget Price Tag
Google just crashed the AI party again — and this time, it brought receipts. Gemini 3.8 Flash officially landed September 2, with Google promising stronger coding, reasoning and autonomous agents. Its third Flash release in six weeks? Somebody in Mountain View apparently lost the snooze button.
According to Google's launch announcement, the upgrade keeps 3.7 Flash's introductory token prices while pushing harder on complicated tasks. That's an attention-grabbing pitch for developers watching their AI bills.
The Benchmarks: Flash Comes Out Swinging
Here's a selection from Google's published benchmark comparison. These are Google-reported results, not independent testing by this publication. Higher scores are better on the measures below.
| Benchmark | Gemini 3.8 Flash | Gemini 3.7 Flash | Claude Opus 5 | GPT-5.6 Sol |
|---|---|---|---|---|
| DeepSWE v1.1 | 73.7% | 65.3% | 74.0% | 72.7% |
| Terminal-bench 2.1 | 89.4% | 85.8% | 89.1% | 88.8% |
| Terminal-bench 4.0 | 19.1% | 11.2% | 51.8% | 37.3% |
| Vals Finance Agent v2 | 61.4% | 59.0% | 58.6% | 53.8% |
| Harvey's Legal Agent Benchmark — all-pass rate | 10.0% | 8.8% | 6.7% | 2.5% |
| HLE-Verified | 54.9% | 53.6% | 54.4% | 54.5% |
The coding headline: Flash climbs 8.4 percentage points over its predecessor on DeepSWE, finishing just 0.3 points behind Opus 5. That's a close finish on this software-engineering test — although a tiny score gap alone doesn't establish equivalent performance.
Hold the Victory Champagne
Terminal-bench 4.0 delivers the plot twist. Flash improves to 19.1%, but Opus 5 scores 51.8% and GPT-5.6 Sol reaches 37.3%. Google has a competitive contender here; the scoreboard doesn't support a clean sweep. Different benchmark versions test different task sets, so that 89.4% on version 2.1 cannot be treated as interchangeable with its 4.0 result.
Meanwhile, Google's model page highlights wins in finance, legal workflows and expert reasoning. The legal result deserves its label: 10% is an all-pass rate on that benchmark, not a blanket measure of legal accuracy.
The Price Tag Has Fine Print
Introductory API pricing is $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026. Google's announced rates rise to $1.50 and $7.50, respectively, on January 1, 2027.
And watch the meter: Google says difficult tasks can consume more tokens. An unchanged token price doesn't guarantee an unchanged bill. Developers can adjust effort levels to manage that tradeoff.
Where You Can Get It
Google lists access through the Gemini API, AI Studio, Antigravity and Gemini Enterprise. Consumer access includes Google AI Pro and Ultra subscribers across the Gemini app, Search's AI Mode and Gemini in Google Sheets. The separately announced 3.8 Flash Cyber is offered to trusted defenders through Google's Fairwind Program. See Google's availability and pricing details.
Flash has earned a spot in the audition. Whether it gets the starring role depends on how it handles your actual workload — and what the bill looks like when the curtain falls.
Comments
No comments yet. Be the first to comment!
Leave a Comment