Google launched Gemini 3.7 Flash on August 13, 2026, solely three weeks after Gemini 3.6 Flash. The brand new mannequin delivers substantial positive factors in coding, doc evaluation and workflow automation whereas chopping the introductory API value to $0.75 per million enter tokens and $3.75 per million output tokens.
These enhancements are actual. Gemini 3.7 Flash climbed from 49.0% to 65.3% on DeepSWE v1.1 and reached 30.4% on Zapier’s AutomationBench, up from 17.0% for its predecessor. Impartial testing from Synthetic Evaluation additionally positioned it among the many quickest reasoning fashions at roughly 340 output tokens per second.
The launch headline leaves out a number of situations. The discounted value expires on the finish of 2026, different fashions stay cheaper, and the buyer rollout via Gemini Spark excludes a number of giant markets. Google’s mannequin additionally loses to rivals on some agent and computer-use evaluations. Builders subsequently have to separate a robust Flash improve from the broader declare that it’s the finest or most cost-effective mannequin for each workload.
What Is Gemini 3.7 Flash
Gemini 3.7 Flash is Google’s newest high-speed mannequin for coding, brokers and data work. It replaces Gemini 3.6 Flash as the corporate’s predominant workhorse mannequin solely three weeks after the earlier model arrived.
Google describes the replace as the results of developer suggestions and algorithmic enhancements. Gemini 3.7 Flash is designed to comply with directions extra reliably and get well from roadblocks throughout multi-step duties.
MemeBurn reported Gemini 3.7 Flash references appearing in Google’s Python SDK someday earlier than the official launch. Google has now confirmed entry via the Gemini API, Google AI Studio, Android Studio, Antigravity and Gemini Enterprise.
For particular person customers, Gemini 3.7 Flash additionally powers Gemini Spark for Google AI Professional and Extremely subscribers in supported nations. That shopper route has regional restrictions that don’t essentially apply to developer API entry.
The Benchmark Beneficial properties Are Substantial
Google’s official Gemini 3.7 Flash announcement exhibits constant enchancment over Gemini 3.6 Flash throughout coding, internet improvement, doc evaluation and automation.
| Benchmark | Gemini 3.7 Flash | Gemini 3.6 Flash | Change |
| DeepSWE v1.1 | 65.3% | 49.0% | Up 16.3 factors |
| FrontierCode 1.1 Major | 43.6% | 34.4% | Up 9.2 factors |
| WebDev Area Elo | 1,588 | 1,538 | Up 50 factors |
| GDP.pdf | 34.0% | 22.0% | Up 12 factors |
| AutomationBench | 30.4% | 17.0% | Up 13.4 factors |
The DeepSWE enchancment issues for coding brokers as a result of the benchmark covers longer software program engineering duties. FrontierCode signifies stronger first-pass code high quality, whereas WebDev Area factors to raised internet interfaces.
Synthetic Evaluation gave Gemini 3.7 Flash Excessive an Intelligence Index rating of 56, 4 factors above Gemini 3.6 Flash. It additionally measured output at round 340 tokens per second, almost thrice the velocity of GPT-5.6 Terra in its comparability. Manufacturing groups ought to nonetheless measure activity success, retries and whole token use somewhat than extrapolating from velocity.
The Half-Value Supply Ends in December
Google has set Gemini 3.7 Flash at $0.75 per million enter tokens and $3.75 per million output tokens, half the unique value of Gemini 3.6 Flash.
Google’s introductory price expires on December 31, 2026. Starting January 1, 2027, enter pricing will increase to $1.50 and output pricing rises to $7.50.
The supply is clearly disclosed, however budgets ought to use the 2027 price for functions anticipated to stay in manufacturing subsequent yr.
On the time of writing on August 14, Gemini 3.7 Flash can be not the lowest-priced main API mannequin. OpenAI lists GPT-5.6 Luna at $0.20 for enter and $1.20 for output. MemeBurn has coated how the GPT-5.6 Luna price cut changed the market for high-volume AI workloads.
DeepSeek lists V4 Flash at $0.14 for uncached enter and $0.28 for output, making its present output price greater than 13 occasions decrease. New DeepSeek V4 pricing takes impact on August 17, so the hole might quickly slim.
Value comparisons additionally have to account for functionality. Luna and V4 Flash goal high-volume effectivity, whereas Gemini 3.7 Flash is competing on stronger coding, multimodal and agent efficiency. The related determine is price per profitable activity, not the most affordable token in isolation.
Sturdy Outcomes Do Not Make It the Greatest at All the things
Gemini 3.7 Flash performs effectively throughout a broad benchmark set, however Google’s personal comparability doesn’t present a common lead.
TechTimes reported that GPT-5.6 Terra scored 50.2% on OSWorld 2.0, forward of Gemini 3.7 Flash at 38.1%. Claude Sonnet 5 additionally led on Agent’s Final Examination with 33.3%, in contrast with Gemini at 26.3%. Synthetic Evaluation positioned Gemini at 56 on its Intelligence Index, one level behind GPT-5.6 Terra and Muse Spark 1.2.
OSWorld measures interplay with pc environments, whereas Agent’s Final Examination checks tough multi-step reasoning. Gemini’s coding and velocity benefits don’t assure the identical rating in these duties.
AutomationBench wants comparable context. The benchmark is built by Zapier and checks end-to-end workflows throughout 47 instruments in gross sales, advertising, operations, help, finance and human assets. Gemini 3.7 Flash presently leads its public leaderboard at 30.44%.
The duties replicate patterns throughout Zapier’s platform, making the outcome helpful for enterprise automation. It stays one surroundings, and groups utilizing totally different instruments, permissions and information may even see one other rating.
Gemini Spark Is Not Out there Throughout A lot of Europe
Google says Gemini 3.7 Flash will energy Gemini Spark for Google AI Professional and Extremely subscribers. Spark can handle multi-step duties, work with related apps and proceed sure jobs within the background.
The Gemini Apps support page states that Spark is unavailable within the European Financial Space, the UK, Switzerland and Nigeria. Customers additionally must be a minimum of 18, use a private Google account and preserve Gemini Apps exercise enabled.
This restriction applies to Spark somewhat than the mannequin’s complete availability. Builders might produce other routes via Google AI Studio, the Gemini API or enterprise merchandise.
Google says Spark reaches greater than 160 nations, however the excluded markets nonetheless cowl a big share of European customers.
Who Ought to Use Gemini 3.7 Flash
Gemini 3.7 Flash fits coding brokers, doc processing and high-throughput internet functions, particularly the place latency and first-pass code high quality matter.
Enterprise trials ought to file activity success, response time, token use, retries and human assessment, then recalculate the outcome utilizing the January 2027 value.
Groups selecting totally on API price have cheaper choices. GPT-5.6 Luna and DeepSeek V4 Flash presently undercut Gemini, though their functionality profiles differ. MemeBurn’s broader GPT-5.6 model comparison explains the place OpenAI’s Sol, Terra and Luna tiers match.
Customers within the EEA, Britain, Switzerland and Nigeria can’t use Gemini Spark at launch.
Gemini 3.7 Flash is subsequently a robust workhorse launch with unusually quick output and significant coding positive factors. Its worth is dependent upon whether or not these benefits survive a crew’s manufacturing workload after the introductory low cost disappears.
FAQs
How A lot Does Gemini 3.7 Flash Value
Gemini 3.7 Flash prices $0.75 per million enter tokens and $3.75 per million output tokens via December 31, 2026. The charges rise to $1.50 for enter and $7.50 for output on January 1, 2027.
Is Gemini 3.7 Flash Higher Than Claude Sonnet 5
It is dependent upon the duty. Gemini 3.7 Flash leads on Zapier’s AutomationBench and provides quicker, cheaper inference. Claude Sonnet 5 scored larger on Agent’s Final Examination within the printed comparability, indicating a bonus on that tough multi-step agent take a look at.
Is Gemini 3.7 Flash the Most cost-effective AI Mannequin
No. As of August 14, 2026, GPT-5.6 Luna and DeepSeek V4 Flash have decrease printed enter and output costs. DeepSeek has introduced that its V4 API pricing will change on August 17.
The place Is Gemini 3.7 Flash Out there
Builders can entry the mannequin via the Gemini API, Google AI Studio, Android Studio and Antigravity. It is usually accessible in enterprise merchandise. Shopper entry via Gemini Spark excludes the EEA, the UK, Switzerland and Nigeria at launch.
