DeepSeek · The Reasoning-Model Era
DeepSeek R1
The open-weights reasoning model that printed an industry shockwave. Trained at a fraction of frontier-lab costs.
By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026
R1's release was the first time an open-weights model matched the frontier reasoning benchmarks at a fraction of the training cost. The technical report described a training pipeline simple enough that a hundred labs reproduced it within six weeks. Subsequent open-weights releases now treat R1's recipe as the floor. The model itself is now used primarily as a backbone for fine-tuning rather than direct deployment.
Field signature
Returns long <think> blocks before answers. Reasoning visible.
Specifications
| Released | 2025-01-20 |
|---|---|
| Context window | 64,000 tokens |
| Pricing | Free if self-hosted; ~$0.55/$2.19 per million tokens on inference services |
| Modalities | text |
| License | MIT (with use-case carve-outs) |
| Era | The Reasoning-Model Era |
Strengths
- Cost
- Open weights
- Competitive reasoning benchmarks
Weaknesses
- Older underlying base model (V3)
- Refusal behavior is unpredictable
- China-origin data filtering visible on some prompts
Authentication markers
The fingerprints by which DeepSeek R1 can be identified from its output alone.
| Tell | Meaning |
|---|---|
| <think> blocks before final answer. | DeepSeek R1 family. |
| Refusal patterns include 'I am sorry, I cannot…' verbatim. | DeepSeek training artefact. |
Notable works
- Triggering the open-weights reasoning wave of Q1 2025
Market position
$0.55 - $2.19 per million tokens via Together / Fireworks / Groq
Partner offer
Partner offerings listed for operator convenience. See disclosure for terms.
View partner →Affiliate link — see disclosure.
Primary sources
- [1] DeepSeek: Models
From the Almanac shop
The Operator's Compendium
Every agent harness, every routing pattern, every cost trick. 90-page PDF.
$29 — Coming soon