Doubao 1.5 Vision Pro (32k context) โ extended-context vision-language variant.
Doubao-1.5-vision-pro-32k is ByteDance's flagship vision-language model released alongside Doubao 1.5 Pro in January 2025 (model id suffix 250115). It extends the sparse MoE architecture of the 1.5 Pro family with comprehensive upgrades in multimodal data processing, dynamic resolution handling, and fine-grained visual-text alignment, enabling strong performance on text-rich image understanding and visual reasoning tasks.
For TheRouter operators, Doubao-1.5-vision-pro-32k offers a cost-efficient Chinese-first multimodal endpoint that supports standard OpenAI vision chat format. It excels at document QA, chart/table extraction, OCR-heavy workflows, and GUI understanding at dramatically lower price than GPT-4o or Claude 3.5 Sonnet vision variants. The model is fully reachable through api.therouter.ai/v1 with the same production inference stack (PD-disaggregation, W4A8 quantization) used by the text-only 1.5 Pro sibling.
| Type | Rate |
|---|---|
| Input | $0.486 / 1M tokens |
| Output | $1.46 / 1M tokens |
| Release date | 2025-01-15www.aibase.com โ | verified |
| Architecture | Sparse MoE vision-language model (extension of Doubao-1.5-pro)team.doubao.com โ | verified |
| Input modalities | Text + Image (no native video)www.volcengine.com โ | verified |
| License | Closed (API only) | verified |
| Training cutoff | Not publicly disclosed | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
DocVQA Strong text-rich document understanding; one of the highest reported scores among 2025 MLLMs. | 96.7%% | arxiv.org โ | |
InfoVQA Excellent infographic and chart reasoning performance. | 89.3%% | arxiv.org โ | |
VisuLogic Highest score among compared models on this visual reasoning benchmark (still 23+ points below human). | 28.1%% | arxiv.org โ | |
ChartQA Competitive chart and table extraction; exact numeric scores reported in OCR-Reasoning paper. | Strong | arxiv.org โ |
Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.
curl https://api.therouter.ai/v1/chat/completions -H "Content-Type: application/json" -H "Authorization: Bearer $THE_ROUTER_API_KEY" -d '{
"model": "doubao/doubao-1-5-vision-pro-32k",
"messages": [
{"role": "user", "content": "Summarize the key points from this input."}
]
}'curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model":"doubao/doubao-1-5-vision-pro-32k",
"messages":[{"role":"user","content":[
{"type":"text","text":"่ฏท็จไธญๆๆ่ฟฐ่ฟๅผ ๅพ็ไธญ็่กจๆ ผๅ
ๅฎนใ"},
{"type":"image_url","image_url":{"url":"https://therouter.ai/assets/vision-sample.png"}}
]}]
}'ByteDance launched the Doubao visual understanding model Doubao-1.5-vision-pro with comprehensive upgrades in multimodal data processing and dynamic resolution. The model targets text-rich image reasoning and Chinese document workflows at the same aggressive pricing as the text-only 1.5 Pro sibling.
Yes. It accepts the same chat.completions payload structure with image_url content blocks as other OpenAI-compatible vision models exposed through TheRouter.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | www.aibase.com โ | 2026-05-26 | verified |
| Architecture | team.doubao.com โ | 2026-05-26 | verified |
| Input modalities | www.volcengine.com โ | 2026-05-26 | verified |
| License | โ | โ | verified |
| Training cutoff | โ | โ | unknown |
| DocVQA | arxiv.org โ | 2026-05-26 | verified |
| InfoVQA | arxiv.org โ | 2026-05-26 | verified |
| VisuLogic | arxiv.org โ | 2026-05-26 | verified |
| ChartQA | arxiv.org โ | 2026-05-26 | to verify |
| ByteDance releases Doubao-1.5-vision-pro alongside Doubao 1.5 Pro | aibase.com โ | 2026-05-26 | verified |
| Does Doubao-1.5-vision-pro-32k support the standard OpenAI image_url format? | src โ | 2026-05-26 | to verify |