Doubao Seed 1.6 Flash β ultra-low-latency variant of Seed 1.6, ideal for chat and agent loops.
Doubao Seed 1.6 Flash is the ultra-low-latency member of ByteDance's Seed1.6 family, released June 2025 through Volcano Engine Ark. It uses the same 23B-active / 230B-total sparse MoE backbone as the base Seed1.6, but is post-trained and served for minimal time-to-first-token and high token throughput, targeting real-time chat, agent loops, and lightweight tool workflows.
On TheRouter, the model is exposed with a 128K context window and full OpenAI-compatible tool calling. It is the recommended entry point when operators need a fast, cost-efficient Chinese-language model for high-QPS chat or agent traffic where the heavier Seed1.6 or Seed1.6-Thinking variants would add unnecessary latency.
| Type | Rate |
|---|---|
| Input | $0.081 / 1M tokens |
| Output | $0.324 / 1M tokens |
| Release date | 2025-06 (Seed1.6 family)seed.bytedance.com β | verified |
| Architecture | 23B active / 230B total sparse MoEseed.bytedance.com β | verified |
| Context window (TheRouter) | 128K tokensseed.bytedance.com β | verified |
| License / availability | Proprietary (Volcengine Ark API) | verified |
| Training cutoff | Not publicly disclosed | unknown |
| Notable technique | AdaCoT (Adaptive Chain-of-Thought) β FullCoT / NoCoT / AdaCoT modesseed.bytedance.com β | verified |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
C-Eval / CMMLU class Seed1.6 family leads or matches GPT-4.5-class Chinese benchmarks; Flash variant preserves most capability at lower latency/cost. | Competitive at Flash tier | seed.bytedance.com β | |
Latency (TPOT) Third-party measurements on Volcengine Ark report ~10 ms time-per-output-token for the Flash variant. | ~10 msms | aisharenet.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-flash",
"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-flash","messages":[{"role":"user","content":"η¨δΈζ解ι AdaCoT ηδΈη§ζ¨‘εΌγ"}]}'Seed1.6 (23B active / 230B total MoE) launched with multimodal pre-training, 256K context, and Adaptive Chain-of-Thought (AdaCoT) supporting Full/No/Ada modes. Flash variant targets low-latency production use.
Flash is post-trained and served for minimal latency and high throughput. It preserves most reasoning capability of the base model while targeting real-time chat and agent workloads. Use the base Seed 1.6 when you need deeper reasoning or longer CoT traces.
The Flash variant exposed via TheRouter is text-only. For vision-language workloads use doubao/doubao-seed-1-6-vision.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | seed.bytedance.com β | 2026-05-25 | verified |
| Architecture | seed.bytedance.com β | 2026-05-25 | verified |
| Context window (TheRouter) | seed.bytedance.com β | 2026-05-25 | verified |
| License / availability | β | β | verified |
| Training cutoff | β | β | unknown |
| Notable technique | seed.bytedance.com β | 2026-05-25 | verified |
| C-Eval / CMMLU class | seed.bytedance.com β | 2026-05-25 | to verify |
| Latency (TPOT) | aisharenet.com β | 2026-05-25 | to verify |
| ByteDance Seed releases Seed1.6 family with AdaCoT on Volcengine | seed.bytedance.com β | 2026-05-25 | verified |
| How does Seed 1.6 Flash differ from the regular Seed 1.6? | aisharenet.com β | 2026-05-25 | to verify |
| Does Seed 1.6 Flash support vision or video inputs? | seed.bytedance.com β | 2026-05-25 | to verify |