Reference
Full table
Every measured limit for 54 model and provider pairs, measured 2026-10-31. Hover a value for the passing and failing request sizes and the provider's error text. Click a column header to sort.
54 pairs
| Model | Provider | Images | 768×432 frames | Tokens / frame | Inline body | Images by URL | Video file | Video tok/min | Frames/min | Notes |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash (1017) DeepSeek | Fireworks via OpenRouter deepseek/deepseek-v4.1-flash-1017 | 120 | 112 | 223 | ≥ 32 MB | 64 MB | No video | No | — | |
| DeepSeek V4.1 Flash (1017) DeepSeek | Fireworks accounts/fireworks/models/deepseek-v4p1-flash-1017 | 120 | 112 | 223 | ≥ 32 MB | 80 MB | No video | No | — | 1 note |
| DeepSeek V4.1 Flash (1017) DeepSeek | Jarvislabs deepseek-v4.1-flash-1017 | 999 | ≥ 384 | 223 | 12 MB | ≥ 128 MB | No video | No | — | 1 note |
| DeepSeek V4.1 Flash (1017) DeepSeek | Modal deepseek-ai/DeepSeek-V4.1-Flash-1017 | 999 | ≥ 384 | 223 | ≥ 32 MB | ≥ 128 MB | No video | No | — | |
| DeepSeek V4.1 Flash (1017) DeepSeek | Together AI deepseek-ai/DeepSeek-V4.1-Flash-1017 | 839 | ≥ 384 | 223 | ≥ 32 MB | ≥ 128 MB | No video | No | — | strict tool order 1 note |
| Gemini 3.7 Flash Google | Google AI Studio via OpenRouter google/gemini-3.7-flash | 960 | ≥ 384 | 1,100 | 14 MB | 14 MB | 12 MB | 3,961 | 59.8 | |
| Gemini 3.7 Flash Google | Google AI Studio (flex) via OpenRouter google/gemini-3.7-flash | 896 | — | — | — | — | — | 3,961 | 59.8 | |
| Gemini 3.7 Flash Google | Google AI Studio (priority) via OpenRouter google/gemini-3.7-flash | 960 | — | — | — | — | — | 3,961 | 59.8 | |
| Gemini 3.7 Flash Google | Google Vertex (global) via OpenRouter google/gemini-3.7-flash | 768 | — | — | — | — | — | 3,960 | 60 | |
| Gemini 3.7 Flash Google | Google Vertex (global, flex) via OpenRouter google/gemini-3.7-flash | 960 | — | — | — | — | — | 3,960 | 60 | |
| Gemini 3.7 Flash Google | Google Vertex (global, priority) via OpenRouter google/gemini-3.7-flash | ≥ 512? | — | — | — | — | — | 3,960 | 60 | |
| Gemini 3.7 Flash Google | Google Vertex (us) via OpenRouter google/gemini-3.7-flash | 768 | — | — | — | — | — | 3,960 | 60 | |
| Gemma 4 31B Google | Jarvislabs gemma-4-31b-it | 4 | 4 | 1,110 | 12 MB | ≥ 128 MB | 256 MB | 467 | 6.2 ⁎ | 1 note |
| GLM 5.3 Flash Z.ai | Fireworks accounts/fireworks/models/glm-5p3-flash | 120 | 112 | 450 | ≥ 32 MB | 80 MB | No video | 46,994 | — | 1 note |
| GLM 5.3 Flash Z.ai | Jarvislabs glm-5.3-flash | ≥ 1,024 | 128 | 450 | 12 MB | ≥ 128 MB | 64 MB | ? | — | |
| GLM 5.3 Flash Z.ai | Together AI zai-org/GLM-5.3-Flash | 1,004 | ≥ 384 | 450 | ≥ 32 MB | ≥ 128 MB | 256 MB | 952 | — | strict tool order 1 note |
| GLM 5.3 Flash Z.ai | Z.ai via OpenRouter z-ai/glm-5.3-flash | 724 | ≥ 384 | 450 | ≥ 32 MB | ≥ 128 MB | No video | No | — | rejects response_format 1 note |
| Kimi K3 MoonshotAI | Fireworks accounts/fireworks/models/kimi-k3 | 120 | 112 | 457 | ≥ 32 MB | 80 MB | 96 MB | ? | — | rejects usage, reasoning 1 note |
| Kimi K3 MoonshotAI | Modal moonshotai/Kimi-K3 | ≥ 1,024 | ≥ 384 | 457 | ≥ 32 MB | ≥ 128 MB | No video | No | — | |
| Kimi K3 MoonshotAI | Together AI moonshotai/Kimi-K3 | ≥ 512? | ≥ 384 | 457 | ≥ 32 MB | ≥ 64 MB? | ≥ 512 MB | Ignored | — | strict tool order 2 notes |
| Muse Spark (open) Meta | Fireworks accounts/fireworks/models/muse-spark-open | 120 | 112 | 407 | ≥ 32 MB | 14 MB | No video | No | — | rejects usage, reasoning 1 note |
| Qwen3.6 35B A3B Qwen | AkashML (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | 2,499 | 12.5 ⁎ | |
| Qwen3.6 35B A3B Qwen | Atlas Cloud (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | 35,757 | 118.4 | |
| Qwen3.6 35B A3B Qwen | Auto-routed via OpenRouter qwen/qwen3.6-35b-a3b | 170 | 160 | 338 | ≥ 32 MB | ≥ 128 MB | ≥ 512 MB | 2,274 | 5.7 ⁎ | |
| Qwen3.6 35B A3B Qwen | CoreWeave (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | 2,274 | — | |
| Qwen3.6 35B A3B Qwen | DarkBloom (fp4) via OpenRouter qwen/qwen3.6-35b-a3b | 16 | — | — | — | — | — | No | — | |
| Qwen3.6 35B A3B Qwen | DeepInfra (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 30 | — | — | — | — | — | 639 | 27 ⁎ | |
| Qwen3.6 35B A3B Qwen | DekaLLM via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | 2,330 | — | |
| Qwen3.6 35B A3B Qwen | Parasail (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | Rejected | — | |
| Qwen3.6 35B A3B Qwen | Phala via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | No | — | |
| Qwen3.6 35B A3B Qwen | SiliconFlow (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 170 | — | — | — | — | — | 281 | 0.5 ⁎ | |
| Qwen3.6 35B A3B Qwen | Venice (fp8) via OpenRouter qwen/qwen3.6-35b-a3b | 10 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | AkashML (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | 2,499 | 6.3 ⁎ | |
| Qwen3.8 27B Qwen | Alibaba Cloud via OpenRouter qwen/qwen3.8-27b | 250 | 224 | 338 | 20 MB | 28 MB | ≥ 512 MB | 35,640 | 60 | rejects response_format 1 note |
| Qwen3.8 27B Qwen | Cerebras (fp16) via OpenRouter qwen/qwen3.8-27b | 2 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | Chutes (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | 2,499 | 4.4 ⁎ | |
| Qwen3.8 27B Qwen | Cloudflare via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | CoreWeave (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | 2,274 | — | |
| Qwen3.8 27B Qwen | DarkBloom (fp4) via OpenRouter qwen/qwen3.8-27b | 64 | — | — | — | — | — | Rejected | — | |
| Qwen3.8 27B Qwen | DeepInfra (bf16) via OpenRouter qwen/qwen3.8-27b | 30 | — | — | — | — | — | 2,499 | 6.3 ⁎ | |
| Qwen3.8 27B Qwen | DekaLLM via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | 2,330 | 4 ⁎ | |
| Qwen3.8 27B Qwen | IonStream (fp8) via OpenRouter qwen/qwen3.8-27b | 8 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | Jarvislabs qwen-3.8-27b-fp8 | 4 | 4 | 355 | 12 MB | ≥ 128 MB | 256 MB | 2,499 | 6.3 ⁎ | 1 note |
| Qwen3.8 27B Qwen | Mancer (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | Rejected | — | |
| Qwen3.8 27B Qwen | Modal via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | ModelRun (fp4) via OpenRouter qwen/qwen3.8-27b | 16 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | NEAR AI (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | Novita via OpenRouter qwen/qwen3.8-27b | 250 | — | — | — | — | — | 35,640 | — | |
| Qwen3.8 27B Qwen | Parasail (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | 2,499 | 6.3 ⁎ | |
| Qwen3.8 27B Qwen | Phala via OpenRouter qwen/qwen3.8-27b | ? | — | — | — | — | — | 2,499 | 6 ⁎ | |
| Qwen3.8 27B Qwen | Reka (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | 2,499 | 6.3 ⁎ | |
| Qwen3.8 27B Qwen | Tenstorrent (fp4) via OpenRouter qwen/qwen3.8-27b | ? | — | — | — | — | — | No | — | |
| Qwen3.8 27B Qwen | Venice (fp8) via OpenRouter qwen/qwen3.8-27b | 10 | — | — | — | — | — | 2,499 | 6.3 ⁎ | |
| Qwen3.8 27B Qwen | Wafer (fp8) via OpenRouter qwen/qwen3.8-27b | 170 | — | — | — | — | — | No | — |
No pairs match. Clear the filter to see all of them.
120 measured limit ≥ 16 MB no failure up to our probe's cap ? search stopped on a timeout 170 tight for video work 4 very tight No video not accepted
Reading the numbers
- Images and 768×432 frames are different tests. Tiny images measure the count limit alone; real-size frames hit pixel, payload or context-window limits first on some providers.
- Inline body and images by URL often have different limits on the same endpoint. If one is tight, try the other.
- A value is the largest request that passed. Leave headroom: limits change, and some depend on your plan.