Doubao 1.5 Pro β ByteDance flagship general-purpose chat model with tools and JSON mode.
Doubao-1.5-pro-32k is ByteDance's flagship sparse MoE language model released January 2025. It uses a highly sparse Mixture-of-Experts architecture that activates only a fraction of its parameters while delivering performance comparable to much larger dense models (7Γ performance leverage over equivalent dense training). The 32k variant exposed via TheRouter supports tools, JSON mode, and a "Deep Thinking" reasoning mode.
For TheRouter operators, Doubao-1.5-pro-32k is a cost-efficient Chinese-first reasoning model that matches or exceeds GPT-4o/Claude 3.5 Sonnet on many Chinese and reasoning benchmarks at ~50Γ lower price than GPT-4o. It is fully OpenAI-compatible through api.therouter.ai/v1 and benefits from ByteDance's production-grade inference stack (PD-disaggregation, W4A8 quantization, speculative decoding). The model is the direct predecessor of the Seed 1.6 family.
| Type | Rate |
|---|---|
| Input | $0.1296 / 1M tokens |
| Output | $0.324 / 1M tokens |
| Release date | 2025-01-22www.marktechpost.com β | verified |
| Architecture | Sparse MoE (7Γ performance leverage vs dense equivalent)team.doubao.com β | verified |
| License | Closed (API only) | verified |
| Training cutoff | Not publicly disclosed | unknown |
| Benchmark | Distribution | Score | Source |
|---|---|---|---|
AIME Interim preview version (AS1) surpassed o1-preview and o1 on AIME; final model further improved via RL. | Surpasses o1-preview | team.doubao.com β | |
Reasoning (multiple) Matches or exceeds GPT-4o and Claude 3.5 Sonnet on many reasoning and Chinese-knowledge benchmarks at dramatically lower cost. | Matches GPT-4o / Claude 3.5 Sonnet | www.marktechpost.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-1-5-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-pro-32k","messages":[{"role":"user","content":"θ―·η¨δΈζ解ι MoE θ΄θ½½εθ‘‘γ"}]}'ByteDance launched Doubao-1.5-pro, a sparse MoE model that surpasses o1-preview on AIME in preview and matches GPT-4o/Claude 3.5 Sonnet on many benchmarks at ~50Γ lower cost. The model emphasizes training-inference co-design and a fully autonomous data production pipeline.
Yes. It supports the standard OpenAI tools / tool_choice interface and is listed with tools in standard-models.yaml.
| source | URL | retrieved | |
|---|---|---|---|
| Release date | www.marktechpost.com β | 2026-05-25 | verified |
| Architecture | team.doubao.com β | 2026-05-25 | verified |
| License | β | β | verified |
| Training cutoff | β | β | unknown |
| AIME | team.doubao.com β | 2026-05-25 | to verify |
| Reasoning (multiple) | www.marktechpost.com β | 2026-05-25 | to verify |
| ByteDance releases Doubao-1.5-pro with Deep Thinking mode | team.doubao.com β | 2026-05-25 | verified |
| Does Doubao-1.5-pro-32k support tool calling? | src β | 2026-05-25 | to verify |