Google’s AI coding platform Antigravity has introduced a new Gemini 3.5 Flash Low model designed to reduce token usage and improve efficiency for developers working on lightweight software engineering tasks.

The update marks the latest expansion of Antigravity, the DeepMind backed AI coding platform launched in late 2025 as competition intensifies across the rapidly growing AI assisted software development market.

Antigravity lead builder Varun Mohan announced the rollout this week, revealing that Gemini 3.5 Flash Low generates roughly 45 percent fewer tokens compared to the platform’s existing Gemini 3.5 Flash Medium model while still outperforming the older Gemini 3 Flash system on several software engineering benchmarks.

The move comes after users repeatedly raised concerns about high token consumption, API costs, and resource efficiency while using larger coding models for simpler development tasks.

According to Mohan, the new Flash Low model is optimized for lightweight coding workflows, debugging, autocomplete assistance, and basic engineering operations where developers may not require the heavier reasoning capabilities of larger AI systems.

The company also confirmed that quotas have been reset across all paid Antigravity plans following the update.

The rollout quickly generated discussion across developer communities, particularly among users looking for lower cost AI coding solutions as model usage expenses continue rising across the industry.

Several developers praised the platform’s rapid response to community feedback regarding token efficiency and practical workflow optimization.

AI builder Anshul Ramachandran described the new model as useful for handling routine coding tasks while reserving more advanced models for complex engineering problems, allowing teams to scale AI usage more efficiently across projects.

The launch reflects a broader shift happening inside the AI industry as companies increasingly focus not only on model intelligence, but also on operational efficiency, inference costs, speed, and scalability.

Over the past year, AI coding platforms have rapidly evolved from simple autocomplete assistants into fully integrated software engineering systems capable of writing code, debugging applications, analyzing repositories, generating tests, and handling multi step development workflows.

However, growing usage has also exposed one of the industry’s biggest challenges: computational cost.

Larger reasoning models often consume massive token volumes even for relatively simple development tasks, creating higher infrastructure costs for both providers and users. This has pushed companies toward introducing smaller, faster, and more resource efficient AI models designed for everyday workflows.

Gemini 3.5 Flash Low appears to target exactly that segment.

The model joins Antigravity’s growing lineup of Flash Medium and Flash High systems, giving developers more flexibility to balance performance, speed, and token consumption depending on workload complexity.

Some users, however, also questioned whether the lower token model could introduce tradeoffs in reasoning quality, image handling capabilities, or advanced coding performance during larger workflows.

The discussion highlights an increasingly important trend across the AI sector where efficiency is becoming nearly as valuable as raw intelligence.

As AI coding adoption accelerates across startups, enterprise software teams, and independent developers, platforms are now competing heavily on pricing efficiency, latency reduction, workflow optimization, and scalable deployment rather than benchmark scores alone.

For Google and DeepMind, the Antigravity expansion represents another strategic move to strengthen Gemini’s position inside the growing AI developer ecosystem, where rivals including OpenAI, Anthropic, GitHub Copilot, and Cursor continue competing aggressively for developer adoption.

The addition of Gemini 3.5 Flash Low also signals how AI coding tools are entering a new phase focused on practical daily usability, not just maximum capability, as developers increasingly demand systems that are faster, cheaper, and easier to scale across real world software projects.