DeepSeek V3.2 Exp Think Now Live on TheRouter — Aggressive Sparse-Attention Reasoning, CN-Region Routing
DeepSeek V3.2 Exp Think pairs the V3.2 base architecture's Sparse Attention with an explicit reasoning trace, landing at $0.56/M input and $0.84/M output — the cheapest reasoning-capable model in our catalog. Served via the new deepseek route (CN-region partner capacity) for customers who want CN-region latency or sourcing diversification.
Why this model
- Sparse Attention keeps long-context inference cheap without quality loss vs the dense V3.x branch.
- Explicit thinking mode exposes the reasoning chain in
completion_tokens_details.reasoning_tokens, billed at the standard output rate. - 128K context with a hard ceiling on per-token cost — no tiered pricing surprises.
- CN-region sourcing via our partner network gives mainland-China deployments sub-100ms round-trip to upstream.
Pricing & capabilities
| Property | Value |
|---|---|
| Standard alias | deepseek/deepseek-v3.2-exp-think |
| Input price | $0.56 / 1M tokens |
| Output price | $0.84 / 1M tokens |
| Context | 128,000 tokens |
| Capabilities | Streaming · Reasoning · JSON mode |
| Served via | deepseek (CN-region partner capacity) |
Prices shown are launch-time figures; see each model page for the current billed rate.
Try it
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-v3.2-exp-think",
"messages": [{"role": "user", "content": "Prove that sqrt(2) is irrational."}],
"provider": {"only": ["deepseek"]}
}'The provider.only filter is optional — without it, the router picks the cheapest healthy route automatically. Drop it to let cost-aware routing pick between our CN-region partner network and any future direct DeepSeek provider.
See the models reference for the full catalog, or read the blog for more launch notes.