Gemini 3.7 Flash Doubled Its Workflow Score
Ignore the coding benchmarks for a second. The number that matters here is the one about workflows finishing.
Founder, Automation Squad ·
The short answer
Gemini 3.7 Flash launched August 13, 2026 at $0.75 per million input tokens and $3.75 output, half its standard rate, through December 31, 2026. On AutomationBench, which tests real business workflows, it scores 30.4% against 3.6 Flash's 17.0% — a near doubling, though it still means most of those workflows do not complete.
The numbers, the price cliff, and a ten-run test worth doing
Use the table to decide whether to look, and the test below to decide whether to switch. Budget about half an hour.
Pick the automation you gave up on, not the one that works
Specifically a multi-step job that kept losing the thread halfway through. A single-prompt task will not tell you anything new.
Rebuild just that workflow against Gemini 3.7 Flash in AI Studio
Do not port your whole stack. One workflow, same inputs, same success criteria as before.
Run it ten times on real inputs and count completions
Ten is enough to separate a real improvement from a lucky first run. Count only runs that finish end to end without you stepping in.
Write down the failure mode of every run that did not finish
If eight of ten fail the same way, that is a fixable prompt problem, not a model ceiling. If they fail differently each time, the task is still beyond it.
If it clears your bar, diary the January 1 price change now
Anything you cost-model before January is priced at half what it will cost afterwards. Put the real number in the plan while it is in front of you.
| Benchmark | 3.7 Flash | 3.6 Flash | What it tests |
|---|---|---|---|
| AutomationBench | 30.4% | 17.0% | Real-world business workflows |
| DeepSWE v1.1 | 65.3% | 49.0% | Software engineering |
| FrontierCode 1.1 Main | 43.6% | 34.4% | Production code |
| GDP.pdf | 34.0% | 22.0% | Document comprehension |
| WebDev Arena (Elo) | 1588 | 1538 | Web development |
| Price in / out, to Dec 31 2026 | $0.75 / $3.75 | — | Per million tokens |
| Price in / out, from Jan 1 2027 | $1.50 / $7.50 | — | Introductory rate doubles |
Take it with you
GEMINI 3.7 FLASH — the ten-run test
Source: https://automationsquad.com/news/gemini-3-7-flash-automation-benchmark/
BENCHMARK 3.7 FLASH 3.6 FLASH TESTS
AutomationBench 30.4% 17.0% Real business workflows
DeepSWE v1.1 65.3% 49.0% Software engineering
FrontierCode 1.1 43.6% 34.4% Production code
GDP.pdf 34.0% 22.0% Document comprehension
WebDev Arena (Elo) 1588 1538 Web development
PRICE PER MILLION TOKENS
Aug 13 - Dec 31 2026: $0.75 in / $3.75 out (introductory)
From Jan 1 2027: $1.50 in / $7.50 out (doubles)
THE TEST
[ ] Pick the MULTI-STEP automation I abandoned (not one that already works)
[ ] Rebuild only that workflow against Gemini 3.7 Flash in AI Studio
[ ] Run it 10x on real inputs
[ ] Count runs that finish end to end with NO intervention: ___ / 10
[ ] Note the failure mode of each run that didn't finish
- same failure every time -> fixable prompt problem
- different failure each time -> still beyond the model
[ ] If it clears my bar, diary the Jan 1 2027 price doubling now
Read 30.4% honestly: it's a near doubling AND most workflows still don't
complete. Worth testing on a task you gave up on; not worth trusting unattended.Google released Gemini 3.7 Flash on August 13, 2026, three weeks after 3.6 Flash, with introductory pricing that runs until the end of the year.
The facts, from Google's launch post: introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, doubling to $1.50 and $7.50 on January 1, 2027. Against 3.6 Flash, it scores 30.4% versus 17.0% on AutomationBench, which tests real-world business workflows; 65.3% versus 49.0% on DeepSWE v1.1; 43.6% versus 34.4% on FrontierCode 1.1 Main; 34.0% versus 22.0% on GDP.pdf for document comprehension; and 1588 versus 1538 Elo on WebDev Arena. It is available through Google AI Studio, Android Studio and Google Antigravity for developers, the Gemini Enterprise Agent Platform for companies, and Gemini Spark for AI Pro and Ultra subscribers across more than 160 countries.
Automation Squad's take: the coding scores will get the coverage and the AutomationBench number is the one worth your time. A jump from 17.0% to 30.4% on real business workflows is not a model getting cleverer at puzzles — it is a model failing to finish a multi-step job somewhat less often. That said, read the number honestly in both directions: 30.4% still means the majority of those workflows do not complete. This is a model that has become worth testing on a task you previously gave up on, not a model you should trust unattended. The introductory price is the other half of the story, and it has an expiry date on it — anything you build a cost model around before January is priced at half what it will cost afterwards.
Run this now: find the automation you abandoned because the model kept losing the thread halfway through — the multi-step one, not the single-prompt one. Rebuild just that workflow against Gemini 3.7 Flash in AI Studio and run it ten times on real inputs, counting how many complete end to end without you intervening. Ten runs is enough to tell a real improvement from a lucky one. If it clears your bar, note the January 1 price change in your budget now, while the number is in front of you.
By the numbers
The data behind the story — figures from cited sources, with our own analysis labelled.
AutomationBench is the closest of these to the question most readers actually have: can it run my workflow without me?
- Aug 13 – Dec 31, 2026Introductory pricing$0.75 in / $3.75 out
- From Jan 1, 2027Standard pricing$1.50 in / $7.50 out
Prices are per million tokens. The introductory rate is exactly half the standard rate.
Questions people are asking
- How much does Gemini 3.7 Flash cost?
- $0.75 per million input tokens and $3.75 per million output tokens as an introductory rate through December 31, 2026. On January 1, 2027 it doubles to $1.50 and $7.50. Any cost model built before January is working from half the eventual price.
- What is AutomationBench and why does it matter more than the coding scores?
- It tests real-world business workflows rather than isolated puzzles, which makes it the closest published proxy for 'can this model run my multi-step process without me'. Gemini 3.7 Flash scores 30.4% against 3.6 Flash's 17.0%.
- Is Gemini 3.7 Flash good enough to run workflows unattended?
- No. A 30.4% score on AutomationBench is a large improvement and still means the majority of those workflows do not complete. Treat it as a reason to retest something you previously abandoned, not as a reason to remove human checkpoints.
- Where can I use Gemini 3.7 Flash?
- Google AI Studio, Android Studio and Google Antigravity for developers; the Gemini Enterprise Agent Platform and Gemini Enterprise app for companies; and Gemini Spark for Google AI Pro and Ultra subscribers across more than 160 countries.
Sources
Further reading
- Google — Gemini 3.7 Flash: our most intelligent workhorse model — The full benchmark table, including the ones we did not pull out here.
Last checked August 15, 2026 against the primary sources above, by Automation Squad Research. Spot an error? [email protected].