Gemini 3.8 Flash

Artificial Analysis gave Gemini 3.7 Flash an intelligence index of 56, which is 4 points higher than 3.6 Flash.

The test covered four vision models: Claude Sonnet 5 (Anthropic), Gemini 3 Flash (Google), Gemini 3.1 Flash-Lite (Google), and GPT-5.4 nano (OpenAI).

Some users found in tests with 3 coding prompts that completion times were consistently in the 30–45 second range, with 3.8 Flash being at most about 5 seconds faster.

The Gemini Flash series faces direct competition from OpenAI's GPT-4o-mini and Anthropic's Claude Haiku.

In the per-answer cost including caching, Claude Haiku 5.5 was the lowest at $0.03, Claude Sonnet 5.5 was $0.35, and Gemini 3.8 Flash was the highest at $0.79.

In the speed test, Claude Sonnet and Haiku were the fastest and most stable, never exceeding 3 minutes; with Mistral and Gemini, about one in every four questions took over 10 minutes, and some approached 25 minutes.

Per-token price does not equal per-answer price: Gemini and Mistral have cheap per-token prices, but with an average of as many as 38 steps per question, their per-answer cost is comparable to or even higher than the currently used model.

3.8 Flash is reliable at programmatic processing of structured file formats (.xlsx, .docx, complex CSV parsers), with almost zero syntax errors in cell formatting, dynamic formula evaluation, and multi-sheet merging.

3.8 Flash performs better in IDEs like Antigravity than in bare chat interfaces, because the IDE provides real-time lint feedback, file tree inspection, and terminal test execution as external compensation.

Some enterprise developers report that 3.8 Flash has failure modes such as fabricating APIs, scope creep, and infinite reasoning loops that exhaust the quota.

When lacking visual references, 3.8 Flash produces conservative Web styling (gray buttons, #0000ff links, Bootstrap-style layouts), but after providing modern UI screenshots or a design token system, its CSS generation quality improves substantially.

3.8 Flash often introduces regressions when performing large monolithic refactors involving multiple concerns, missing adjacent functions (such as the cart badge update listener) and breaking the error popup handler; the solution is to split the task across multiple files and execute it step by step.

In Antigravity, a ten-step agent loop takes 7 to 10 minutes under 3.1 Pro and completes in under 90 seconds under 3.8 Flash.

In actual testing on the POS reporting automation project, Gemini 3.1 Pro had a single-step latency of 28–45 seconds, consumed about 42% of quota, and failed after 3 refactoring iterations; Gemini 3.8 Flash High had a single-step latency of 3–6 seconds, consumed about 11% of quota, and succeeded on the first try.

According to a benchmark run headlessly via Antigravity on the same seven-constraint article planning task (three trials each, n=6, measured on 2026-09-27), Gemini 3.8 Flash (High) and Claude Opus 4.6 (Thinking) both scored 17/17 on machine-verified constraint adherence, but the median wall-clock time was 23 seconds for Gemini versus 67 seconds for Opus.

On two-step reasoning questions, Claude Haiku 4.5 scored 6/30, GPT-5.4 mini scored 3/30, while Claude Sonnet 5.5, Gemini 3.8 Flash, and Gemma 4 all scored 30/30.

On the Intelligence Index, Haiku 5.5 surpassed Gemini 3.8 Flash and DeepSeek 4.1.

On the unified agent setup leaderboard actually operated by DataCurve, three models are tied for first place at 74%, including Google's Gemini 3.8 Flash, while Argon is not listed.

Google's Claude Code competitor, Anti-Gravity CLI, is free to use with quotas that refresh weekly and runs models such as Gemini 3.8 Flash, but does not include Argon.

On Cursor Bench, Sonnet 5.5 consumed 271,920 tokens per task, surpassing Opus's 218k and Gemini 3.8 Flash's 162k.