Summary of Key Points
Google DeepMind recently released three new lightweight AI models, featuring fast speed and low cost. However, the lack of improvement in intelligence has led to mixed reviews. The launch of the flagship model Gemini 3.5 Pro has been postponed, and Google’s existing models have dropped out of the top ten global intelligence rankings. Google’s AI strategy prioritizes “speed and cost” over model capabilities, which has sparked controversy among developers. As a result, Google’s stock price has declined, and its market value has been surpassed by Apple.
1. Three New Models: Faster, Cheaper, but Not Smarter
The three new models have different focuses:
- Gemini 3.6 Flash (the main model): Uses fewer computing resources, allowing it to perform higher-quality tasks with the same investment, yet its intelligence has not improved.
- 3.5 Flash-Lite: Designed for everyday tasks such as document processing and searching, offering both speed and cost-effectiveness.
- 3.5 Flash Cyber: Specialized in cybersecurity, focusing on identifying software vulnerabilities.
Third-party evaluations show that these models have not become more intelligent. For example, the 3.6 Flash ranked 12th in front-end code testing, with an intelligence score of 50 (the same as its predecessor), but it is 50% faster and slightly cheaper. In other words, Google has optimized efficiency without enhancing cognitive abilities.
2. Poor Intelligence Performance: Dropping Out of the Top Ten
A significant contrast compared to last year:
- When Google released Gemini 3 in November, it was on par with OpenAI; now, none of Google’s models are in the top ten of intelligence rankings.
- Leading models include Anthropic, OpenAI, as well as domestic ones like Kimi K3 and Zhipu GLM-5.2.
- Developers complain that the new models perform poorly on tasks involving front-end development and spatial reasoning, with some comparing Google to the “rabbit” in the Aesop’s fable of “The Tortoise and the Hare”—always falling asleep after gaining a temporary lead.
This is a stark contrast to last year, when Gemini 3 drove a surge in Google’s stock price.
3. Postponed Launch of the Flagship Model
The much-anticipated Gemini 3.5 Pro was supposed to be released one month after the I/O conference in May, but it is still missing.
- Media reports suggest that the release has been delayed due to internal tests not meeting performance expectations.
- A Google technical official stated that they are still testing with partners, without providing a specific timeline.
- Interestingly, Google has mentioned the start of pre-training for the next generation, Gemini 4, raising speculation about whether resources have been shifted towards this new model.
4. Strategic Controversy: Business Smarts or Short-Sightedness?
Google’s AI strategy has divided opinions:
- Supporters: Google has a large ecosystem with products like Android, Chrome, and Workspace. By integrating AI into these services (e.g., search and office software), it is being strategic rather than chasing the “smartest” models at all costs. Many small companies fail due to excessive spending; by saving money, Google can sustain its business longer.
- Critics: They argue that intelligence is the core of an AI company. If model capabilities stagnate, even successful commercialization will lose developer trust and affect future market prospects. Some emphasize that innovation requires depth, not just speed.
5. Market Reaction: Stock Price Decline, Market Value Drop
Google’s performance is reflected in its stock market:
- On July 21, Google’s stock price fell by more than 1%, closing at $347.
- Since mid-May, the stock has lost nearly 15% in value.
- Its market value has dropped from second to third place, with Apple now surpassing it (Google’s current market value is around $4.2 trillion).
The market clearly does not approve of Google’s current AI strategy; investors value technological leadership over mere cost optimization.
In Conclusion
Google’s current approach focuses on generating revenue first before pursuing greater intelligence. However, this may harm its technical reputation in the short term. Whether it will affect the long-term health of its ecosystem depends on its ability to balance “speed” and “intelligence.”