Multimodal LLM comparison — text, image, audio

Multimodal LLM comparison starts with the modality matrix, not with Elo. If the model cannot accept an image, it is not in the running. If it can, you still have to price image tokens and test the actual screenshots.

Model detail pages list modality badges from the OpenRouter catalog. Compare two vision-capable rows on price and quality index, then run a handful of images yourself — text arena scores will not rank OCR.

Popular frontier models by quality

Live OpenRouter pricing and Arena quality, cached about an hour. Not a static blog table.

ModelProviderBlend $ / 1MQualityContext
Claude Opus 5Anthropic15.0093.41,000,000
Claude Fable 5Anthropic30.0091.11,000,000
Qwen3.8 27BQwen1.7189.01,000,000
Gemini 3.7 FlashGoogle2.2588.41,048,576
DeepSeek V4 Pro 0423DeepSeek1.1488.21,024,000

Input vs output modalities

Many models take images and emit text. Fewer emit images. Audio and video support is spottier and billed differently. Read both compare-table rows (modalities in / out) before you design a voice or video feature around a text SKU.

Quality is task-shaped

Screenshot QA, receipt OCR, and meme jokes are different jobs. Keep separate eval sets. A Gemini row might win documents while a GPT row wins UI screenshots this quarter — and reverse next quarter.

Pipelines vs native multimodal

ASR + text LLM + TTS can beat a native audio model on cost and control. Native multimodal wins when you need joint reasoning (“the speaker points at this chart”). Don’t pay VLM prices for a problem Whisper plus GPT-mini solves.

Leaderboard limits

Our quality index is text Elo. Speed weights are not vision-latency. Use this site to shortlist on price and flags, then measure multimodal quality off-site or in your own harness.

Suggested compare set

One OpenAI, one Google, one Anthropic vision SKU, and one cheap open VLM if you have a policy to include open weights. Four columns on /compare is enough.

FAQ

Which model is best at images?

It depends on the image type. Shortlist multimodal rows here, then test your images.

Can I compare modalities on this site?

Yes — the compare table lists input and output modality strings from the catalog.

Are video models included?

If they appear in the OpenRouter text-capable catalog we ingest. Many video endpoints do not.

Does Arena Elo include vision?

Our quality index uses the text leaderboard. Don’t treat it as a VLM ranking.
multimodal LLM comparisonvision language model comparisonGPT-4o vs Gemini visionbest multimodal AI model
Browse the leaderboard

Related guides

Also try: LLM leaderboard, compare tool, live arena.