Google rolled out a new lightweight artificial intelligence (AI) model specialized for coding and agent work, but the release of its top-tier model, which has been silent for two months, was delayed again. As the launch of the top-tier "Pro" model drags on, Google's AI model competitiveness appears to be falling behind SpaceXAI, once considered a latecomer.
Google said on the 13th (local time) that it will introduce the lightweight AI model "Gemini 3.7 Flash" to corporations and paying individual customers. It comes just three weeks after the release of "Gemini 3.6 Flash" on the 21st of last month.
According to Google, the new model's coding and bug-fixing capabilities have improved significantly, enabling it to generate production-ready code that can run immediately in a single attempt.
In fact, Gemini 3.7 Flash's key performance indicators (benchmarks) for coding and more were similar to, or in some areas ahead of, midrange models such as Anthropic's Claude Sonnet 5 and OpenAI's GPT-5.6 Terra. Considering that its application programming interface (API) usage price is two to three times cheaper than those models, its price-to-performance ratio is high.
It especially showed strength in response speed. According to AI model analytics firm Artificial Analysis (AA), Gemini 3.7 Flash's response speed is 340 tokens per second, the second fastest after Gemini 3.5 Flash Lite (358 tokens per second). Among major competing models, none exceeded 200 tokens per second.
However, as a lightweight model, it did not rank at the top for overall intelligence performance. AA rated Gemini 3.7 Flash's intelligence index (AAII) at 56 points, placing it ninth between GPT-5.6 Terra (57 points) and DeepSeek V4 Pro 0813 (53 points).
The problem is that Google's top-tier model release keeps getting delayed. Google has yet to unveil "Gemini 3.5 Pro," which was initially slated for release in June, even after two months have passed.
Meanwhile, rivals have been rolling out high-performance models in quick succession, pulling ahead of Google. "Grok 4.6," unveiled by SpaceXAI on the 12th, scored 61 on AAII, just one point behind Claude Fable 5 (62 points), and tied for third with GPT-5.6 Sol.
Meta's "Muse Spark 1.2" (57 points), as well as China's Kimi K3 (60 points) and Qwen 3.8 Max (58 points), also received higher marks than Gemini 3.7 Flash.
While it is showing competitiveness among lightweight models with an edge in price and speed, concerns are growing that Google is losing its lead in the high-performance AI model race as the top-tier model release is delayed.