Apology to Gemini 3.8 Flash: Antigravity CLI Unlocks Its True Potential

Gemini 3.8 Flash in Antigravity CLI

After previous disappointing tests on the Gemini web interface and desktop client, the author resolved an eligibility issue for Antigravity and tested Gemini 3.8 Flash inside Antigravity CLI. In the CLI environment, the model excels across writing quality, aesthetic judgment, multimodal understanding, response speed, and quota generosity. The author explicitly warns against using the web version, labeling it "weak" in a screenshot caption.

Comparison with Claude Opus 4.6

The author notes that Claude Opus 4.6 offers comparable writing capability but suffers from a severe quota restriction: after roughly three prompts, a five‑hour cooldown is triggered, rendering it nearly unusable for sustained work. They attribute this to the model being billed via standard API pricing, which is inherently expensive. A screenshot from the platform "谷子" (likely a unified AI gateway) shows that Gemini and Claude quotas are tracked separately, with the Gemini allowance being far more durable.

Speculation on Qoder Quota Consumption

The author observes that many users report rapid quota depletion in Qoder, especially when using the international version's 1.6× and 3.2× flagship models. They hypothesize that Qoder routes these requests through AWS, where the higher underlying cost naturally accelerates quota consumption.

Concrete Multimodal Test

As a practical demonstration, the author fed Gemini 3.8 Flash over 460 images from a "pelican cycling test" and asked it to produce a picture scroll and a chronological timeline. The one‑shot result was impressive, though not flawless.