Gemini 3.7 Flash tuo koodaukseen jättiloikan ja mullistavan hinnan – Googlen lippulaivamalli 3.5 Pro on edelleen kadoksissa

On August 13, 2026, Google launched Gemini 3.7 Flash, its latest workhorse AI model designed for coding and agent workflows, just three weeks after Gemini 3.6 Flash. The release comes at a time of significant internal upheaval at Google DeepMind, and the company offered no update on its long-delayed flagship model, Gemini 3.5 Pro .

What Google Announced With Gemini 3.7 Flash

Google positioned Gemini 3.7 Flash as "our most intelligent workhorse model yet for coding and agents" . The model delivers substantial gains across software engineering, web development, and agentic workflows, according to Google's official blog post .

Benchmark Improvements Over 3.6 Flash

Google reported significant jumps across several key benchmarks :

Benchmark	3.6 Flash	3.7 Flash	Improvement
DeepSWE v1.1 (SWE-bench verified)	49.0%	65.3%	+16.3 pts
FrontierCode 1.1 Main	34.4%	43.6%	+9.2 pts
AutomationBench	17.0%	30.4%	+13.4 pts (nearly doubled)
GDP.pdf (grounded document processing)	22.0%	34.0%	+12.0 pts
WebDev Arena Elo	~1538	1588	+50 pts

These figures come from Google's official announcement and have been corroborated by multiple outlets including Reuters, Bloomberg, and tech publications .

Pricing and Availability

The pricing for Gemini 3.7 Flash is notably aggressive :

• Introductory price: $0.75 per million input tokens, $3.75 per million output tokens (through December 31, 2026)
• Post-intro price (January 1, 2027 onward): $1.50 per million input, $7.50 per million output
• Context window: 1 million tokens, multimodal input
• Availability: Generally available (GA) via the Gemini API, AI Studio, and Gemini Spark — the free consumer tier of Gemini — immediately at launch

Google also temporarily cut the price of Gemini 3.6 Flash to match the new 3.7 Flash pricing .

The Status of Gemini 3.5 Pro: Still Delayed, Possibly Canceled

Status: Gemini 3.5 Pro remains delayed with no release date provided. Promised to developers as far back as May 2026 at Google I/O, it has been repeatedly pushed back . On August 13/14, Google offered no update on when or if it will ship . At the time of 3.7 Flash's launch, the Flash line was already two versions ahead of the delayed Pro flagship .

Pricing (unconfirmed — estimates only): Google has not published official API pricing for 3.5 Pro . Based on leaked and estimated reports from analysts and tech publications, the expected pricing is:

• Input: ~$15 per million tokens (estimated)
• Output: ~$60 per million tokens (estimated)
• Context window: 2 million tokens (double most rivals)

These estimates remain unofficial as Google has not released a model card.

Could Google skip Gemini 3.5 Pro entirely? Multiple reports note that the prolonged delay, combined with the leadership upheaval and reports that early Gemini 4 training is already underway, has fueled speculation that Google may scrap or bypass 3.5 Pro and jump directly to a 4.0 flagship . Bloomberg and Reuters both note Google's silence on the matter is unprecedented for a model this far into its delay cycle . The Next Web reported that "Google's cheap model is now two versions ahead of its flagship" .

Competitive Positioning: 3.7 Flash vs Claude Sonnet 5 vs GPT-5.6 Terra

Google claims Gemini 3.7 Flash outperformed comparable models from Anthropic and OpenAI across nine benchmarks . Here is the price-performance comparison :

	Gemini 3.7 Flash (intro)	Claude Sonnet 5 (intro)	GPT-5.6 Terra
Input price	$0.75/M	$2/M (until Aug 31)	$2.50/M
Output price	$3.75/M	$10/M (until Aug 31)	$15/M
DeepSWE v1.1	65.3%	63.2% (SWE-bench Pro)	Not published
Terminal-Bench 2.1	Not published vs rivals	78.4%	87.1%
Context	1M	1M	Large
Status	Live	Live	Live

Key takeaway: Gemini 3.7 Flash is dramatically cheaper than both rivals at its introductory pricing — approximately 2.7x cheaper than Sonnet 5 and 3.3x cheaper than Terra on input — while posting a competitive SWE-bench score (65.3% vs Sonnet 5's 63.2%). However, it does not match Terra's Terminal-Bench performance (87.1%) in that specific agentic coding benchmark .

Market commentary describes 3.7 Flash as offering "performance comparable to Claude Sonnet 5 and GPT-5.6 Terra" at roughly half the price of its predecessor, making it a strong value play for coding and agent workloads .

Leadership Turbulence at Google DeepMind

A major internal shakeup unfolded in the week before the 3.7 Flash launch, adding uncertainty to Google's AI roadmap.

Sergey Brin urging "all in": Google co-founder Sergey Brin has been actively pressing key AI staff to "go all in on Gemini" and close the distance with rivals. Multiple sources say he has been directly involved in pushing researchers to accelerate timelines. In April, he addressed hundreds of employees at a town hall and urged Google DeepMind to move faster .

Demis Hassabis stepping aside: On August 5, 2026, Google announced that Demis Hassabis — DeepMind's co-founder, Nobel Prize recipient, and CEO — would step down from day-to-day operations and take a chairman/chief scientist role, focusing on AGI strategy and drug discovery .

Koray Kavukcuoglu takes over: DeepMind's CTO Koray Kavukcuoglu will now have final say on major DeepMind decisions as the de facto operational leader, reporting directly to CEO Sundar Pichai . The move effectively centralizes Google's AI leadership in Mountain View .

Departure of Gemini's technical co-leads:

• Jeff Dean — Alphabet's chief scientist and one of Google's first employees — left to found a new startup, Discovery Loop, along with three other top Google AI staffers .
• The two original technical co-leads of Gemini departed from Google in June 2026 to join AI rivals .
• Additional exits of key engineers were reported, driven by low morale, missed deadlines, and burnout .

Fortune reported that the leadership exodus was triggered by "stalled models, missed deadlines, and staff burnout" .

Summary of Open Questions

• Gemini 3.5 Pro release: Unclear. Google is silent. Speculation is growing that it may be canceled in favor of a jump to Gemini 4 .
• Gemini Spark integration: Confirmed for immediate availability at launch .
• Gemini 4 training: Early training is reportedly underway, but no official confirmation or timeline has been provided .

Bottom line: Google shipped a strong Flash update that rivals models from Anthropic and OpenAI at a fraction of the price — but it did so amid a leadership crisis. Its flagship Pro model remains stuck in limbo, and the company's talent exodus suggests that its AI roadmap may face further turbulence before a clear successor to 3.5 Pro emerges.