Model
Model explorer

R1-1776

OPEN
Perplexity AI · R1-1776 family · released Feb 18, 2025

MIT post-train of DeepSeek-R1 removing CCP censorship on ~300 topics via 40K-prompt multilingual SFT with NeMo 2.0; not currently listed on Perplexity's hosted API pricing or HF Inference Providers.

ReasoningCodingVisionFunction callingTool useAgentic
1566.7
Elo · unrated
Parameters
671B
Active params
37B (MoE)
Context
128K tokens
Architecture
Mixture-of-Experts (DeepSeek-R1/V3 architecture), 671B total / 37B active
License
MIT
Languages
—
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
AIMEMath79.8%#80
best: GPT-5.2 · 100.0%
Run it locally
VRAM @ Q4
335 GB
VRAM @ FP16
1342 GB
Fits on (Q4)
M3 Ultra 512GB
671B MoE far exceeds a single RTX 4090's 24GB even at Q4 (~335GB needed); community low-bit GGUF (e.g. Unsloth Q2_K_XL) needs multi-GPU/CPU offload.
Quantizations
GGUF
R1-1776 family
Elo progression across releases
R1-1776Feb 20251566.7
Resources
HFperplexity-ai/r1-1776DOCSperplexity.ai/hub/blog/open-sourcing-r1-1776
API price weights · each benchmark row carries its own source badge (see methodology)