AI Image, Video & Audio Model Pricing
47 image, video and audio models — arena Elo rankings from blind preference votes, plus list prices with the page each number came from.
These models are tracked separately from the text leaderboard: they are billed per image, per second or per character rather than per token, so they do not carry an AMR score.
Snapshot 2026-10-28 · updated weekly · 27 of them with a verified list price.
Image
Image models bill per image or per 1M tokens depending on the vendor; the unit is stated in each row.
Ranking: blind preference votes (Elo) from the Artificial Analysis arena, snapshot 2026-10-28. Rows without a rank are listed for pricing only. (12 arena-ranked)
| # | Model | Vendor | Arena Elo | Price |
|---|---|---|---|---|
| 1 | GPT Image 2.5 Sunburst (max) Image generation, top quality tier | OpenAI | 1222 | — |
| 2 | GPT Image 2.5 Flare (max) Image generation, speed-biased tier | OpenAI | 1207 | — |
| 3 | GPT Image 2 (high) Image generation, high effort tier | OpenAI | 1186 | — |
| 4 | Gemini Nano Banana 2.1 Image generation and editing model | 1176 | $1.50 in / $7.50 out per 1M tokens | |
| 5 | MAI-Image-2.6 Image generation | Microsoft | 1166 | — |
| 6 | Grok Imagine Image 2.0 Image generation | xAI | 1166 | — |
| 7 | Gemini 3.1 Flash Image Native image generation, flash tier | 1140 | $0.50 in / $3.00 out per 1M tokens | |
| 8 | Gemini 3 Pro Image Native image generation; image input billed at $2.00 per 1M tokens | 1118 | $2.00 in / $12.00 out per 1M tokens | |
| 9 | FLUX 3 Image Image generation | Black Forest Labs | 1117 | — |
| 10 | Gemini 3.1 Flash Lite Image Native image generation, lite tier | 1107 | $0.25 in / $1.50 out per 1M tokens | |
| 11 | Seedream 5.0 Pro Image generation | ByteDance | 1100 | — |
| 12 | Qwen-Image-3.0-Pro Image generation | Alibaba | 1095 | — |
| — | GLM-Image Text-to-image generation | Zhipu AI | — | $0.015 per image |
| — | CogView-4 Text-to-image generation | Zhipu AI | — | $0.01 per image |
| — | MiniMax image-01 Text-to-image generation | MiniMax | — | $0.0035 per image |
| — | GPT-5.4 Image 2 Image generation and editing; billed per token, not per image | OpenAI | — | $8.00 in / $15.00 out per 1M tokens |
| — | GPT-5 Image Image generation; billed per token | OpenAI | — | $10.00 in / $10.00 out per 1M tokens |
| — | GPT-5 Image Mini Compact image generation tier | OpenAI | — | $2.50 in / $2.00 out per 1M tokens |
Video
Video output is not available through OpenRouter as of this snapshot, so these prices come from vendor pages. Kuaishou and ByteDance list API rates behind their consoles.
Ranking: blind preference votes (Elo) from the Artificial Analysis arena, snapshot 2026-10-28. Rows without a rank are listed for pricing only. (7 arena-ranked)
| # | Model | Vendor | Arena Elo | Price |
|---|---|---|---|---|
| 1 | Wan 3.0 Text-to-video generation | Alibaba | 1174 | — |
| 2 | Utopai X Custom video model, post-trained on MiniMax H3 | Utopai | 1168 | — |
| 3 | MiniMax H3 (768P) 768P resolution | MiniMax | 1158 | $0.08 per second (output) |
| 4 | Dreamina Seedance 2.5 Text-to-video generation | ByteDance | 1156 | — |
| 5 | FLUX 3 Max (720p) Text-to-video generation, max tier | Black Forest Labs | 1155 | — |
| 6 | MiniMax H3 Max Video generation, max quality tier | MiniMax | 1152 | $0.08 per second (output, 768P) / $0.05 (480P) |
| 7 | Seedance 2Unverified Video generation via Volcengine; multiple resolution tiers | ByteDance | 1136 | — |
| — | MiniMax H3 (2K) 2K resolution; input reference video billed at $0.13/s | MiniMax | — | $0.13 per second (output) |
| — | MiniMax H3 Highspeed (768P) 768P, faster generation tier | MiniMax | — | $0.06 per second (output) |
| — | MiniMax H3 Lite (768P) 768P budget tier | MiniMax | — | $0.04 per second (output) |
| — | MiniMax H3 Lite (480P) 480P budget tier | MiniMax | — | $0.025 per second (output) |
| — | CogVideoX-3 Text-to-video generation | Zhipu AI | — | $0.2 per video |
| — | Kling 4.0Unverified Up to 30-second native videos, 4K with 10-bit HDR, up to 10 reference images and 5 clips per generation | Kuaishou | — | — |
| — | VeoUnverified Video generation through the Gemini API | — | — |
Audio & Music
Speech and music: text-to-speech is billed per character, recognition per hour or per token, music per song.
Ranking: blind preference votes (Elo) from the Artificial Analysis arena, snapshot 2026-10-28. Rows without a rank are listed for pricing only. (4 arena-ranked)
| # | Model | Vendor | Arena Elo | Price |
|---|---|---|---|---|
| 1 | HeyGen Voice Text-to-speech | HeyGen | 1195 | — |
| 2 | Qwen-Audio-3.1-TTS-Plus Text-to-speech | Alibaba | 1184 | — |
| 3 | Eleven v4 Text-to-speech | ElevenLabs | 1161 | — |
| 4 | Sonic 3.6 Text-to-speech, low-latency | Cartesia | 1151 | — |
| — | MiniMax Speech 2.8 HD Text-to-speech, sync and async | MiniMax | — | $100 per 1M characters |
| — | MiniMax Speech 2.8 Turbo Text-to-speech, sync and async | MiniMax | — | $60 per 1M characters |
| — | MiniMax ASR Speech-to-text | MiniMax | — | $0.38 per hour |
| — | MiniMax Voice Design Custom voice creation | MiniMax | — | $3 per voice (previews billed at $60 per 1M characters) |
| — | MiniMax Rapid Voice Cloning Voice cloning | MiniMax | — | $1.5 per voice |
| — | MiniMax Music 3.0 Music generation | MiniMax | — | $0.15 per up-to-5-minute song |
| — | GLM-ASR-2512 Speech recognition | Zhipu AI | — | $0.03 per 1M tokens |
| — | GPT Audio Audio input and output; billed per token | OpenAI | — | $2.50 in / $10.00 out per 1M tokens |
| — | GPT Audio Mini Compact audio input and output tier | OpenAI | — | $0.60 in / $2.40 out per 1M tokens |
| — | Lyria 3 (Preview) Music generation | — | $0 (preview) | |
| — | WhisperUnverified Speech-to-text and translation | OpenAI | — | — |
Multimodal pricing FAQ
Why are these models not on the main leaderboard?
Text models are scored by AMR over written benchmarks. Multimodal models are ranked by arena Elo: blind preference votes collected by Artificial Analysis, shown next to the list price and its source. Where a model has no arena presence it is listed for pricing only, without a rank.
How often are these prices updated?
The snapshot date sits at the top of this page. Vendor pricing pages are checked on the same weekly cycle as the text leaderboard, and each row links to the exact page its number came from so you can verify it yourself.
Why do some rows show no price?
Either the vendor publishes rates only inside a console, or the model has not been through the preference arena yet. Ranked-but-unpriced and priced-but-unranked are both marked with a dash in the missing column rather than a guessed number.