Google Gemini 3.7 Flash: Pricing, Benchmarks, and the Troubled State of Google's AI

On Thursday, August 13, 2026, Google launched Gemini 3.7 Flash, its most intelligent workhorse model yet for coding and agents — while still offering no release date for the long-delayed flagship Gemini 3.5 Pro. The announcement comes alongside a major leadership shakeup at Google DeepMind, with co-founder Demis Hassabis stepping down as CEO.

What Google announced with Gemini 3.7 Flash

Gemini 3.7 Flash was released on August 13, 2026, just three weeks after Gemini 3.6 Flash . It is a refinement of 3.6 Flash with algorithmic improvements to the core reasoning foundation — not a new pretraining run . The model accepts text, images, audio, and video across a 1M-token context window and returns up to 64K output tokens . Google pitches it as the primary agentic workhorse of the Gemini 3 family, bridging the gap between deep-reasoning Pro models and lightweight Flash-Lite models . It is available on Google's APIs, AI Studio, and Antigravity, with a rollout to other products .

Benchmarks vs. its predecessor (Gemini 3.6 Flash)

Benchmark	3.7 Flash	3.6 Flash	Change
AA Intelligence Index	56	52	+4
FrontierCode 1.1 (production code quality)	43.6%	34.4%	+9.2 pp
DeepSWE 1.1 (complex software engineering)	65.3%	48.6%	+16.7 pp
Terminal-Bench 2.1	85.8%	78.0%	+7.8 pp
Terminal-Bench 3.0	14.9%	5.4%	+9.5 pp (2.8x improvement)
AutomationBench	30.4%	17.0%	+13.4 pp
Code Arena WebDev (Elo)	1,588	1,538	+50

The largest gains are in complex software engineering and document understanding. On DeepSWE V1.1, 3.7 Flash's 65% gets close to GPT-5.6 Terra's performance in that specific evaluation .

However, Google's own competitors still lead on widely-cited coding benchmarks. For context: Claude Sonnet 5 scored 76.1% on Terminal-Bench 2.1 and 63.2% on SWE-Bench Pro, and GPT-5.6 Terra scored roughly 70% and 60%, respectively .

Pricing: 50% cheaper than Gemini 3.6 Flash

Gemini 3.7 Flash launched at an introductory price of $0.75 per million input tokens and $3.75 per million output tokens — 50% cheaper than Gemini 3.6 Flash, which was priced at $1.50/$7.50 . This introductory rate runs through December 31, 2026, after which it reverts to $1.50/$7.50 .

Comparative pricing (per million tokens, input/output):

Model	Input ($/MTok)	Output ($/MTok)
Gemini 3.7 Flash (intro)	$0.75	$3.75
Gemini 3.6 Flash	$1.50	$7.50
Claude Sonnet 5 (standard)	$3.00	$15.00
GPT-5.6 Terra	~$2.50	~$15.00

3.7 Flash undercuts both Claude Sonnet 5 and GPT-5.6 Terra by roughly 60–75% on a blended cost basis . But on raw coding benchmarks, the competitors still lead. For developers, 3.7 Flash offers a strong cost-to-performance ratio for high-volume agentic tasks where absolute top-tier accuracy is not required.

Status of the delayed flagship Gemini 3.5 Pro

• May 2026: Announced at Google I/O with a promised June release window . That deadline passed without a release.
• July 2026: Bloomberg reported the model is "months behind schedule" because its coding performance fell short of internal goals . Alphabet's market cap dropped roughly $200 billion (4.4%) on the news .
• As of August 13, 2026: Google still provided no release date or timeline for Gemini 3.5 Pro . The company only says the model is "being tested with partners and coming soon" . It has no entry in the Gemini API and no official spec sheet .

Leadership changes and strategic pressures at Google DeepMind

On August 5, 2026, Alphabet announced a sweeping AI leadership restructuring :

• Demis Hassabis stepped down as Google DeepMind CEO to become chairman of the unit and chief scientist at Alphabet . The move formalized a shift away from day-to-day management toward AGI strategy and scientific research . Tension between Hassabis's long-term scientific ambitions and Google's need to commercialize AI quickly was cited as a factor .
• Jeff Dean (Google's longtime AI leader) and other key engineers left to found a new startup . Two star researchers had departed to AI rivals in June .
• Google is consolidating AI efforts in its California headquarters, weakening the UK-based DeepMind operation .
• Fortune reported the shakeup was driven by low morale, a talent exodus, and model delays — with stalled models, missed deadlines, and staff burnout cited as key contributors . Experts described it as the biggest restructuring since Google Brain merged into DeepMind .

Hassabis was replaced in daily operations by Koray Kavukcuoglu, DeepMind's chief technology officer, who is now senior vice president reporting directly to Sundar Pichai .