Author(s): Caspar Bannink – AI Engineer

Originally published on Towards AI.

Google Is Falling Even Further Behind In The AI Race

A new model, Gemini 3.6 Flash, which cuts paid output pricing from $9.00 to $7.50 per million tokens versus Gemini 3.5 Flash, while its faster as well, Artificial Analysis measures 275.5 output tokens per second instead of 175.7. But is that worth anything?

The author argues that while Gemini 3.6 Flash improves speed and lowers output cost compared to Gemini 3.5 Flash, it doesn’t show evidence of better long-horizon repo-level coding performance: the Artificial Analysis composite intelligence score stays unchanged (50), so the release mostly improves economics for high-volume or short/bounded agent steps. They contrast this with Flash-Lite’s cheaper but lower intelligence score, and claim that what builders really need is “coding-agent receipt”—public proof that the model can reliably inspect the right files, maintain constraints across tool calls, produce robust patches, run meaningful tests, and recover from failures. The piece also critiques Google’s decision-making for routing builders to the right model and suggests that, without reproducible long-running coding benchmarks and completed-work results, developers shouldn’t upgrade their “serious coding agent” default. Finally, the author provides a practical routing rule: use 3.6 Flash for high-volume bounded tasks that benefit from speed and lower output price, use 3.5 Flash-Lite when throughput matters most, and treat long-running agentic coding improvements as unproven until Google supplies stronger public evidence.

Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI