- Google launched Gemini 3.7 Flash on August 13, 2026, just 24 days after 3.6.
- Intro price divided by two : 0.75 / $3.75 per million tokens until December 31.
- It beats Claude Sonnet 5 and GPT-5.6 on production code, but lags behind on long tasks.
- Shadow on the board: Gemini 3.5 Prothe real flagship, is still absent.
Google is accelerating at an unprecedented pace. Just twenty-four days after Gemini 3.6 Flash, the company released 3.7 Flash on August 13, half the price and designed for code and agents. A signal of vitality, but which poorly masks the persistent absence of the flagship model, Gemini 3.5 Pro.
01 · The launch
A “workhorse” at a knockdown price
Google touts Gemini 3.7 Flash as its “smartest working model for code and agents.” The real argument is the price: 0.75 dollars for entry and 3.75 for exit per million tokens, i.e. half as much than 3.6 released three weeks earlier. This promo runs until December 31, 2026; then the price rises to 1.50/7.50 dollars. In terms of speed, the model displays 340 tokens per second, at the top of a ranking of 186 models.
-50%price vs Gemini 3.6
24 daysbetween 3.6 and 3.7
340tokens/s, 1st out of 186 models
43.6%about FrontierCode 1.1 (code)
02 · Benchmarks
Better on code, but not everywhere
Google claims a model that beats its rivals on code, and it’s true… in part. On FrontierCode 1.1 (production code), 3.7 Flash is slightly ahead of Claude Sonnet 5 and GPT-5.6. But on long engineering tasks (DeepSWE) and computer use (OSWorld), GPT-5.6 dominates. In other words, excellent value for money on current code, not an absolute king.
| BenchmarkA standardized test that measures the performance of models on a task (e.g. SWE-bench for coding, MMLU for general knowledge). | Gemini 3.7 Flash | Claude Sonnet 5 | GPT-5.6 |
|---|---|---|---|
| FrontierCode 1.1 (code) | 43.6% | 42.7% | 41.3% |
| DeepSWE (long tasks) | 65.3% | 53.8% | 69.6% |
| OSWorld 2.0 (computer) | 38.1% | , | 50.2% |
| In/out price ($/M) | 0.75 / 3.75 | 2 / 10 | 2 / 12 |
03 · The dark side of the picture
Gemini 3.5 Pro, the Arlesian
Gemini 3.5 Pro delays have led investors to question Google’s roadmap, particularly on code.According to TradingKey / Seeking Alpha
Behind the frantic pace on Flash models hides a gaping hole: the flagship Gemini 3.5 Pro, expected since May, is still not there. Postponed at least three times, it suffered from weaknesses in the code, to the point that Google would have restarted a complete training. Increasing the number of small, fast models is also a way of occupying the field while waiting. In this contextThe context window: the maximum number of tokens that a model can process in a single request (eg: 200k, 1M). tense, Google has appointed Koray Kavukcuoglu as head of its AI, with the mission of catching up with OpenAI and Anthropic.
04 · The race
The same day, OpenAI draws ultra-fast
The race leaves no respite. Also on August 13, OpenAI previewed GPT-5.6 Sol Ultrafast, an up to 14x faster version powered by Cerebras chips, for now by invitation. Two announcements on the same day, both on speed: the sinews of war 2026 is as much latencyThe time elapsed between sending a request and receiving the full response from a model. and price than raw power. To be put into perspective all the same: Gemini’s market share remains debated. The Pangram detector speaks of a collapse from 12% to 1.9%, but this metric only covers text generation; on consumer applications, Gemini is growing at around 25%, behind ChatGPT (45%).
05 · The French issue
What this changes for French devs and SMEs
For a developer or an SME in France, the arithmetic is unstoppable: at $0.75 per million input tokens, Gemini 3.7 Flash costs nearly 70% less than GPT-5.6 for an agent or chatbot. Enough to significantly reduce the cost ofAPIApplication Programming Interface, the programmatic access channel to a service or model.at least until the end of December. The other side of the coin: on the sovereignty side, there is no French equivalent as aggressive on price, Mistral remaining more expensive in this niche. To decide, use our model comparator and our LLM cost calculator. See also our analysis of the Opus 5 vs Gemini 3.6 price war and the Gemini sheet.
Sources
- Google, “Introducing Gemini 3.7 Flash” (blog.google, August 13, 2026)
- VentureBeat, “Gemini 3.7 Flash targets coding and agents with a 50% price cut”
- The Decoder, “Gemini 3.7 Flash lands with coding gains”
- 9to5Google, “Gemini 3.7 Flash launches three weeks after last model”
- OpenAI, “Previewing Ultrafast” (GPT-5.6 Sol, Cerebras)
Frequently asked questions
- When was Gemini 3.7 Flash released?
- Google released Gemini 3.7 Flash on August 13, 2026, just 24 days after Gemini 3.6 Flash. This is an unprecedented pace, which Google attributes to developer feedback and rapid algorithmic gains on its code and agent working models.
- How much does Gemini 3.7 Flash cost?
- The introductory price is $0.75 entry and $3.75 exit per million tokens, half as much as Gemini 3.6, until December 31, 2026. Then, the price rises to $1.50/$7.50 per million tokens.
- Is Gemini 3.7 Flash better than Claude and GPT-5.6?
- On the production code (FrontierCode 1.1), it is slightly ahead of Claude Sonnet 5 and GPT-5.6. But on long engineering tasks (DeepSWE) and computer use (OSWorld), GPT-5.6 dominates. So it’s excellent value for money, not an absolute leader.
- Where is Gemini 3.5 Pro?
- The flagship Gemini 3.5 Pro, expected since May 2026, is still absent. Postponed at least three times, it suffered from weaknesses in the code, to the point that Google would have restarted a complete training. No official release date has been communicated.
- Is Gemini losing market share?
- It depends on the measurement. The Pangram detector suggests a drop from 12% to 1.9%, but only on text generation. On general public applications, Gemini is progressing towards 25%, behind ChatGPT (around 45%) but ahead of Claude.
- Is Gemini 3.7 Flash interesting for a French SME?
- Yes, especially on the cost: at $0.75 per million input tokens, it is around 70% cheaper than GPT-5.6 for an agent or a chatbot, at least until the end of December 2026. There is no French equivalent as aggressive on price to date.
