The arena · AI model review
Gemini 3.5 Flash
by Google
Google's speed-tier flagship: beats Gemini 3.1 Pro on agentic coding at $1.50/$9, 3x pricier than 3 Flash
$9.00/1M out
Paid tier flat rate $1.50/$9.00 per 1M; batch API 50% off ($0.75/$4.50), context caching $0.15/1M plus $1.00/1M per hour storage, free tier available
1M tokens (1,048,576 in / 65,536 out)
$1.50/1M in
$9.00/1M out
text, vision, audio, video, PDF in; text out
No
What is Gemini 3.5 Flash?
Google's fast multimodal reasoning model, released May 19, 2026 at Google I/O. Outperforms Gemini 3.1 Pro on agentic and coding benchmarks (Terminal-Bench 2.1 76.2%, GDPval-AA 1656 Elo) while generating output roughly 4x faster than other frontier models. 1M-token context window with adjustable thinking effort levels.
Gemini 3.5 Flash pros & cons
Pros
- Beats Gemini 3.1 Pro on agentic benchmarks: 76.2% Terminal-Bench 2.1, 1656 Elo GDPval-AA (vs 1314 for 3.1 Pro), 83.6% MCP Atlas
- Artificial Analysis Intelligence Index 50 at high thinking effort, ranking #10 of 170 tracked models
- Very fast generation (~185 output tokens/sec per Artificial Analysis); Google claims 4x faster output than other frontier models
- 1M-token context window (1,048,576) with multimodal input: text, image, audio, video, PDF
- 25% cheaper than Gemini 3.1 Pro ($1.50/$9 vs $2/$12) while outperforming it on production agent workloads
- Adjustable thinking effort (minimal/low/medium/high) plus 50% batch discount and $0.15/1M context caching
Cons
- 3x price increase over Gemini 3 Flash ($0.50/$3.00) and 5-6x over 2.5 Flash; HN devs saw it as Google probing price tolerance
- Over-eager and verbose: devs report it ignores completion criteria and embellishes beyond instructions (compared to Claude's 'Sonnet 3.7 moment')
- Reliability complaints on Flash serving: developers report frequent 503 errors during peak periods
- Weaker on long-horizon agentic tasks with arbitrary tool availability, a recurring theme devs report with Google models
- High time-to-first-token (~23s at high thinking effort per Artificial Analysis), poor fit for latency-sensitive chat
The arena’s verdict
Pick Gemini 3.5 Flash if you run agentic coding or high-volume multimodal pipelines and want near-Pro quality at 4x the speed: it actually beats Gemini 3.1 Pro on Terminal-Bench and GDPval while costing 25% less. Avoid it if you used the Flash line as a budget tier, since it costs 3x its predecessor Gemini 3 Flash, which remains the cheap option at $0.50/$3.00. Also skip it for latency-sensitive chat at high thinking effort (~23s to first token) or strict, no-embellishment output where its verbosity works against you.
Thumbs up or thumbs down
Cast your verdict
Would you recommend Gemini 3.5 Flash, or warn the crowd away?
Top Gemini 3.5 Flash alternatives
All alternativesAnthropic's fastest model: about 90% of Sonnet 4.5's coding skill at $1/$5 per 1M tokens, 200K context.
Anthropic's April 2026 Opus: 87.6% SWE-bench Verified, 1M context, high-res vision, now behind Opus 4.8
Anthropic's flagship Opus-tier model for long-horizon agentic coding; 1M context at $5/$25 per 1M tokens.
Compare Gemini 3.5 Flash head-to-head
Gemini 3.5 Flash: frequently asked questions
How much does Gemini 3.5 Flash cost per 1M tokens?
Gemini 3.5 Flash costs $1.50/1M in per 1M input tokens and $9.00/1M out per 1M output tokens. Paid tier flat rate $1.50/$9.00 per 1M; batch API 50% off ($0.75/$4.50), context caching $0.15/1M plus $1.00/1M per hour storage, free tier available