Doubao Seed 1.6 Vision β vision-language Seed 1.6 variant for multimodal understanding.
Doubao Seed 1.6 Vision (doubao-seed-1.6-vision-250815) is ByteDance's general-purpose multimodal model exposed via Volcano Engine. It belongs to the Seed1.6 series, a sparse MoE family (230B total / 23B active parameters) that incorporates multimodal understanding during pre-training and supports a 256K context window upstream. The Vision variant adds native VisualCoT β the model can crop, rotate, zoom, and annotate images as part of its reasoning chain.
For TheRouter operators, Doubao Seed 1.6 Vision is a cost-effective Chinese-provider vision model with strong GUI-agent, grounding, and video-understanding capabilities. It is fully OpenAI-compatible through api.therouter.ai/v1 and supports temperature, max_tokens, top_p, tools/tool_choice, response_format, and stop β exactly the parameters listed in standard-models.yaml.
| Type | Rate |
|---|---|
| Input | $0.1296 / 1M tokens |
| Output | $1.30 / 1M tokens |
| Release date | August 2025 (250815 snapshot)ohmygpt.com β | verified |
| Architecture | 230B total / 23B active MoE with multimodal pre-trainingseed.bytedance.com β | verified |
| VisualCoT | Native tool-based visual chain-of-thought (crop, rotate, zoom, annotate)developer.volcengine.com β | verified |
| License | Proprietary (Volcano Engine API)www.volcengine.com β | verified |
| Training cutoff | Not publicly disclosed | unknown |
| Inference backends | Volcano Engine (proprietary)www.volcengine.com β | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
MME-SCI (image-only) Science-focused multimodal benchmark; closed-source Doubao-Seed-1.6 outperformed most open VLMs but trailed top Gemini-class models. | 41.32%% | arxiv.org β | |
Gaokao 2025 (multimodal) Seed1.6-Thinking ranked 2nd in science and 1st in humanities among five leading reasoning models on Shandong Gaokao 2025 papers. | 676 (science) / 683 (humanities) | seed.bytedance.com β |
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-seed-1-6-vision",
"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-seed-1-6-vision","messages":[{"role":"user","content":[{"type":"text","text":"ζθΏ°θΏεΌ εΎη"},{"type":"image_url","image_url":{"url":"https://therouter.ai/assets/vision-sample.png"}}]}]}'ByteDance Seed team unveiled the Seed1.6 family with 256K context, multimodal pre-training, AdaCoT adaptive thinking, and strong Gaokao/JEE results. The Vision variant followed in subsequent months.
Volcano Engine documentation and developer evaluations highlight native VisualCoT, 256K context, and strong performance on GUI grounding and video understanding scenarios.
The upstream Volcano Engine endpoint supports 256K. TheRouter exposes the model with the context_length configured in standard-models.yaml (131072 tokens). Use the YAML value for production capacity planning.
VisualCoT is native: the model itself decides when and how to crop/rotate/annotate the image inside its reasoning trace and returns the transformed image as part of the visible CoT. No separate vision-tool microservice is required.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | ohmygpt.com β | 2026-05-25 | verified |
| Architecture | seed.bytedance.com β | 2026-05-25 | verified |
| VisualCoT | developer.volcengine.com β | 2026-05-25 | verified |
| License | www.volcengine.com β | 2026-05-25 | verified |
| Training cutoff | β | β | unknown |
| Inference backends | www.volcengine.com β | 2026-05-25 | verified |
| MME-SCI (image-only) | arxiv.org β | 2026-05-25 | to verify |
| Gaokao 2025 (multimodal) | seed.bytedance.com β | 2026-05-25 | to verify |
| ByteDance announces Seed1.6 series at FORCE conference | seed.bytedance.com β | 2026-05-25 | verified |
| Doubao-Seed-1.6-vision publicly documented (250815) | volcengine.com β | 2026-05-25 | verified |
| Does Doubao Seed 1.6 Vision support the same context window as the upstream 256K model? | www.volcengine.com β | 2026-05-25 | to verify |
| How does VisualCoT differ from standard vision tool calling? | developer.volcengine.com β | 2026-05-25 | to verify |