MiniMax M3 vs Gemini 3.5 Flash: native multimodal vs speed and tools

If you want multimodal input that is fast and cheap, there are two very different options. MiniMax M3 is an open-weight model trained on images and video from the start. Google's Gemini 3.5 Flash is a closed model built for speed.

The short version: MiniMax M3 for open-weight, natively multimodal work, Gemini 3.5 Flash for speed and tool use.

Specs and price

Prices are FastMetal's USD list as of October 25, 2026, rounded to the cent; the current figures are on the models page. Audio input on Gemini 3.5 Flash is billed at $3.30 per million tokens.

On a task of 100,000 input tokens and 10,000 output tokens, that's about $0.04 on MiniMax M3 and $0.26 on Gemini 3.5 Flash. Gemini costs more than eight times as much per output token.

Where each is stronger

MiniMax M3 is an open mixture-of-experts model with 428B total parameters and about 23B active, and it was trained on images and video mixed in from the first stage, not given vision as an add-on later. That makes it a natural fit for agents that look at a screenshot and act on it. It is also strong at coding, with a reported 80.5% on SWE-bench Verified, and its own attention design keeps long contexts fast. Reviewers put it level with GPT-5.5 on the main benchmarks at 5 to 10% of the cost.

Gemini 3.5 Flash wins on speed and tools. It runs at about 278 tokens per second, and it was reported first on multi-step tool use over MCP at 83.6%. It also accepts the widest range of inputs, audio included.

What each gets wrong

MiniMax M3: video eats tokens. Video input gets expensive in tokens quickly, and for text-only work the multimodal training buys you nothing.

Gemini 3.5 Flash: the price went up, and it's not a Pro model. Users complained that output pricing roughly tripled from the previous generation. On real-world code quality it was reported behind Claude Opus 4.7 on SWE-bench Pro. Treat it as a fast model, not a Pro-class one.

Which to use where

• MiniMax M3: agents that work with images, UI automation, and teams that want coding and multimodal input from one open model.
• Gemini 3.5 Flash: tool-heavy pipelines, audio input, and chat where response speed matters most.

Try both with one key

On FastMetal both are behind the same API key and the same balance. Change model and nothing else.

# MiniMax M3
curl https://api.fastmetal.ai/v1/chat/completions \
  -H "Authorization: Bearer $FASTMETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "minimax-m3", "messages": [{"role": "user", "content": "Hello"}]}'

# Gemini 3.5 Flash: same request, different model
curl https://api.fastmetal.ai/v1/chat/completions \
  -H "Authorization: Bearer $FASTMETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemini-3.5-flash", "messages": [{"role": "user", "content": "Hello"}]}'

FAQ

Which is better for multimodal work? Both read images. Gemini 3.5 Flash also takes audio. MiniMax M3 was trained multimodal from the start and is the stronger pick for agents that operate a UI.

Which is cheaper? MiniMax M3, by a wide margin: $0.30 in and $1.16 out per million tokens on FastMetal, against $1.65 and $9.90 for Gemini 3.5 Flash. Current prices are on the models page.

Which is better outside English? We couldn't find a published benchmark that isolates non-English quality for either model. If your work is mostly in another language, test with your own prompts.

Bottom line

MiniMax M3 for open-weight, natively multimodal work, Gemini 3.5 Flash for speed and tool use. They sit in the same fast, cheap, multimodal tier, but they are good at different things. Run both on your own tasks and let that decide.

Both models, with current prices, are in the model catalog.