Multimodal support matrix (major LLMs)
All models support text in/out. The table shows non-text only. Two columns: Input and Output. An icon means the model handles that format. A note means it is conditional (details below). No icon means unsupported. $ are USD / 1M tokens (standard text rates).
Image🔊 Audio🎬 Video PDF📝 DOCX📊 XLSX📑 PPTX
| Model | Input | Output | In $/M | Out $/M |
|---|---|---|---|---|
| Gemini 3.8 Flash GooglePricing · Document | ImageAudioVideoPDFDOCXXLSXPPTXtext extract | — | $0.75→$1.50 | $3.75→$7.50 |
| Gemini 3.7 Flash GooglePricing · Document | ImageAudioVideoPDFDOCXXLSXPPTXtext extract | — | $0.75→$1.50 | $3.75→$7.50 |
| Gemini 3.1 Flash-Lite GooglePricing (audio 2×) | ImageAudioVideoPDFDOCXXLSXPPTXtext extract | — | $0.25 | $1.50 |
| GPT-6 Astra OpenAIModel docs (input: text/image) | Image | Imagevia tool | $10 | $50 |
| GPT-5.6 Sol OpenAIModel docs | Image | Imagevia tool | $4 | $20 |
| GPT-5.6 Luna OpenAIModel docs | Image | Imagevia tool | $0.20 | $1.20 |
| Claude Fable 5.1 AnthropicPDF support | ImagePDF | — | $10 | $50 |
| Claude Mythos 5.1 AnthropicPDF support | ImagePDF | — | $10 | $50 |
| Qwen3.8-Flash AlibabaQwenCloud (image + video understanding) | ImageVideo | — | $0.15 | $0.47 |
| GLM 5.3 Flash Z.aiGLM docs (Video/Image; File is via tools — omitted) | ImageVideo | — | $0.15 | $0.50 |
| Grok 4.6 xAIModel docs (Text, Image) | Image | — | $2 | $6 |
| DeepSeek V4.1 Flash DeepSeekVision · Pricing (V4 Pro had no vision support) | Image | — | $0.15→$0.30 | $0.60→$1.20 |
| MiMo-V2.6-Pro XiaomiModel specs · Pricing | ImageAudioVideo | — | $0.435 | $0.87 |
| MiMo-V2.6-Flash XiaomiModel specs · Pricing | ImageAudioVideo | — | $0.14 | $0.28 |
Support flags verified against primary sources (2026-09-06). This table lists only formats you can pass as native binary to the API (definite yes only). Tool-based text extraction is possible on any model, so it is omitted. Qwen3.8-Flash documents image and video understanding, so those icons are shown. Direct audio/PDF/Office ingest is not documented there, so omitted. Claude PDF is official visual understanding; DOCX/XLSX need conversion first, so omitted.
What the notes mean
- Gemini DOCX / XLSX / PPTX = text extraction. Official Document understanding does visual understanding for PDF only (layout, charts, tables as images). Word / Excel / PowerPoint are ingested as extracted text, not the same visual parse as PDF. You can send a file, but it is not “PDF-equivalent”.
- GPT-6 Astra / GPT-5.6 Sol / Luna image output = via tool. Official model pages list text as the native output. Images require a separate Responses API image generation tool (GPT-Image 2.0) call. The model does not emit pixels itself.
- GLM 5.3 Flash PDF / Office are omitted. Official How to Use documents native input as image_url (images) and video. The File modality and Office deliverables are “call tools, then process/produce documents” — not native binary ingest. Any model can do that with a text-extract tool, so this table treats it as unsupported.
Primary sources (official URLs)
- Google pricing: ai.google.dev/gemini-api/docs/pricing — Gemini 3.8/3.7 Flash intro $0.75/$3.75 (through 2026-12-31), then $1.50/$7.50. 3.1 Flash-Lite is $0.25 for text/image/video and $0.50 for audio.
- Google documents: Document understanding — PDF is visual understanding. Non-PDF is text extraction.
- OpenAI GPT-6 Astra: developers.openai.com/.../gpt-6-astra — Input modalities: text, image. $10 / $50.
- OpenAI GPT-5.6 Sol: gpt-5.6-sol · Luna: gpt-5.6-luna
- Anthropic PDF: PDF support — All active models support PDF. xlsx/docx need conversion first.
- Qwen3.8-Flash: qwencloud.com/models/qwen3.8-flash — “Non-Real-Time Image Understanding” / “Non-Real-Time Video Understanding”. Pricing from the official blog qwen.ai/blog (input $0.15 / output $0.47).
- Z.ai GLM-5.3-Flash: docs.z.ai/.../glm-5.3-flash — How to Use:
image_url. File / Office deliverables are tool-calling (not native binary ingest — omitted). - xAI Grok 4.6: docs.x.ai/.../grok-4.6 — Modalities: Text, Image → Text. $2 / $6. Audio/video are separate APIs.
- DeepSeek Vision: Vision — native image input on
deepseek-flash(V4.1 Flash, since 2026-09-10). The legacydeepseek-v4-flash-vision-expwas folded into V4.1 Flash. V4 Pro continues without vision support. Pricing: Models & Pricing.
Modality-specific billing rules you should know
The same “1M tokens” can price differently by modality or be tokenized. These are the biggest pitfalls.
① Image, video and PDF are tokenized. In Gemini, images, video, and PDFs are tokenized at the standard input rate, with PDFs (DOCUMENT modality) billed at the image-token rate. Video is billed as the input tokens the model actually processes. Video does not scale simply with runtime; the token count varies with the sampled frames and audio the model processes, so the same length can cost differently depending on content and settings. (Google’s video understanding can reduce long-form video input tokens by up to ~88% versus a raw runtime baseline.)
② Some models charge 2x+ for audio input. Gemini 3.1 Flash-Lite ($0.25→audio $0.50), Gemini 3 Flash ($0.50→$1.00), Gemini 2.5 Flash ($0.30→$1.00), and 2.5 Flash-Lite ($0.10→$0.30) bill audio only at 2–3x the text/image/video rate. In contrast, Gemini 3.5 Flash-Lite and 3.6/3.7/3.8 Flash charge audio at the standard rate (the 3.8 official table has no separate modality column). Audio input pricing varies by model and is not a universal rule across all Gemini models. For audio-heavy workloads, model choice matters.
③ Gemini 3.8/3.7/3.6 Flash intro pricing ($0.75/$3.75) runs through 2026-12-31. From 2027-01-01 it rises to $1.50/$7.50. That standard rate applies to image, audio, and video input too. Budget on the standard rate.
④ Live (realtime), image generation, and speech generation (TTS) live in dedicated models. See below. Current core LLMs are “multi-modal in, text out”; to output anything but text you use a dedicated model.
Live streaming / speech / image / video generation are separate models
Core LLMs output text only because each non-text output lives in its own model. Learn them by use case.
| Use case | Example model | In / Out | Pricing (USD) | Source |
|---|---|---|---|---|
| Realtime speech dialogue Live / RealtimeNew | Gemini 3.8 Live Gemini 3.8 Live Extended Thinking Gemini 3.1 Flash Live Preview Gemini 2.5 Flash Native Audio | audio⇄audio | 3.8 Live / 3.8 Live Extended Thinking / 3.1 Flash Live Preview (one shared price table): input text $0.75 · audio $3.00 ($0.005/min) · image/video $1.00 ($0.002/min) output text $4.50 · audio $12.00 ($0.018/min) Output includes thinking tokens | Google pricing · blog |
| Realtime speech dialogue OpenAI Realtime / Audio | gpt-realtime-2.1 gpt-audio-1.5 | audio⇄audio (native) | Realtime: text $4/$24 · audio $32/$64 Audio: text $2.5/$10 · audio $32/$64 | OpenAI Realtime · Audio |
| Image generation / edit | Gemini Nano Banana 2.1 New Gemini 3.1 Flash Image (Nano Banana 2) Gemini 3 Pro Image (Nano Banana Pro) | image→image | Nano Banana 2.1: 1K=$0.0336 · 2K=$0.0504 · 4K=$0.113 per image ($7.50/M text + $30/M images) Flash Image: $60/M (1K=$0.067) · Pro Image: $120/M (1K=$0.134) | |
| Speech generation (TTS)New | Gemini 3.8 Flash TTS Gemini 3.8 Flash-Lite TTS Gemini 3.1 Flash TTS Preview | text→audio | 3.8 Flash TTS: text in $0.50 / audio out $9.00 3.8 Flash-Lite TTS: text in $0.50 / audio out $6.00 Both priced through 2026-12-31; $1.00/$18.00 and $1.00/$12.00 from 2027-01-01 3.1 Flash TTS Preview: text in $1.00 / audio out $20.00 | |
| Video generation | Gemini Omni 1.1 Flash / Veo 3.1 | text/image→video | Omni: video out 5,792 tok/s (720p) ≈$0.10/s · Veo: $0.40/s |
🎤 OpenAI caveat: GPT-6 Astra and the GPT-5.6 line themselves don’t accept audio/video input, but you can do audio dialogue/generation with gpt-realtime-2.1 / gpt-audio-1.5 (native audio in/out) and image generation via the Responses API image tool (GPT-Image 2.0). Pick by “what I feed in” and “what I want out” as a pair, not by a single model name.
Which model, by the in/out you want
A one-line answer for typical multimodal patterns.
Understand an image and answer
Feed audio (conversation)
Gemini 3.8/3.7 Flash take audio at the standard rate. Audio-heavy? Beware 2x billing on 3.1 Flash-Lite
Understand video for summary/analysis
Gemini 3.8 Flash (video+audio). Qwen3.8-Flash also documents Non-Real-Time Video Understanding (no audio input).
Handle Office docs (PDF/PPTX/DOCX/XLSX)
Realtime speech-to-speech
Gemini 3.8 Live / 3.8 Live Extended Thinking (background reasoning) / gpt-realtime-2.1 — all dedicated models
Generate image / audio / video
Gemini Nano Banana 2.1 / 3.8 Flash TTS / Veo 3.1 (all dedicated)
💡 Want the cost? Enter token counts and compare every model’s cost in USD + JPY with the Token Cost Calculator. For multimodal input, estimate using each model’s per-token rate.
All information here was verified on 2026-09-06 against official pricing pages and docs (Google AI · Google Document understanding · OpenAI · Anthropic PDF support · QwenCloud · xAI · DeepSeek Vision · Z.ai). Prices and support change without notice — always re-check the official pages before use.