Back to Models
DeepSeek R1
deepseekdeepseek/deepseek-r1
API guide
Chat + reasoning traces
DeepSeek-R1 returns both the final answer and the full <think> reasoning trace in the response. Use the trace for debugging, auditing, or surfacing intermediate steps to users.
cURL
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-r1",
"messages": [{"role":"user","content":"Solve: 2x+3=17"}]
}'Reasoning traces (<think>)
R1 always emits <think>…</think> blocks before the final answer. Parse reasoning_content (when available) or the raw assistant message to surface the chain-of-thought.
cURL
curl https://api.therouter.ai/v1/chat/completions \
-H "Authorization: Bearer $THEROUTER_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "deepseek/deepseek-r1",
"messages": [{"role":"user","content":"Prove that sqrt(2) is irrational"}]
}'Fact ledger — every claim on this page traces here
| source | URL | retrieved | |
|---|---|---|---|
| Release date | api-docs.deepseek.com ↗ | 2026-05-25 | verified |
| Architecture | github.com ↗ | 2026-05-25 | verified |
| Context length | github.com ↗ | 2026-05-25 | verified |
| License — weights & code | github.com ↗ | 2026-05-25 | verified |
| Nature publication | www.nature.com ↗ | 2026-05-25 | verified |
| Distilled models | github.com ↗ | 2026-05-25 | verified |
| MMLU (Pass@1) | github.com ↗ | 2026-05-25 | verified |
| MATH-500 (Pass@1) | github.com ↗ | 2026-05-25 | verified |
| AIME 2024 (Pass@1) | github.com ↗ | 2026-05-25 | verified |
| GPQA Diamond (Pass@1) | github.com ↗ | 2026-05-25 | verified |
| LiveCodeBench (Pass@1-CoT) | github.com ↗ | 2026-05-25 | verified |
| Codeforces (Percentile) | github.com ↗ | 2026-05-25 | verified |
| ArenaHard (GPT-4-1106) | github.com ↗ | 2026-05-25 | verified |
| DeepSeek-R1 featured on Nature cover — first mainstream LLM to pass rigorous peer review | nature.com ↗ | 2026-05-25 | verified |
| R1-0528 refresh: JSON output, function calling, reduced hallucinations | api-docs.deepseek.com ↗ | 2026-05-25 | verified |
| DeepSeek-R1 and R1-Zero released — pure RL reasoning model open-sourced | api-docs.deepseek.com ↗ | 2026-05-25 | verified |
| How do I access the reasoning trace (<think> block)? | api-docs.deepseek.com ↗ | 2026-05-25 | to verify |
| Is R1 still the recommended model after V4 launch? | api-docs.deepseek.com ↗ | 2026-05-25 | to verify |
| Can I fine-tune on R1 outputs? | github.com ↗ | 2026-05-25 | to verify |