What will be true of DeepSeek's r2 model?
23
1.3kṀ1915Apr 1
92%
Some version of it will score better on GPQA than o1-preview (73% pass@1)
74%
Its base model will be DeepSeek-V3
63%
It will support a longer context length than DeepSeek-r1
55%
DeepSeek will report it scores >= 65% on SWE-Bench Verified
54%
It will score >= 78% on LiveBench when first added (R1 = 71.5% on March 12)
45%
Manifold thinks it is better than expected in a poll asking if it is better or worse than expected
37%
It will have a variant with more parameters than DeepSeek-r1 (not necessarily active)
12%
Reaches Highest Arena Score on Chatbot Arena
everything N/As if a model called R2 created by DeepSeek is not released in 2025
This question is managed and resolved by Manifold.
Get 
1,000 to start trading!
1,000People are also trading
Related questions
Will DeepSeek's next reasoning model be called R3?
4% chance
When will DeepSeek release R2?
When will DeepSeek release V4?
When will Deepseek R2 be released?
6/7/25
Will DeepSeek R2 be open source?
84% chance
Will DeepSeek's next reasoning model be open-sourced?
72% chance
Will I believe in 1 year that DeepSeek R1 was substantially trained via distillation of a US model?
64% chance
Will xAI be ahead of DeepSeek on June 30
65% chance
DeepSeek open-source frontier model after 3/23/25?
82% chance
DeepSeek open-source frontier model after 3/23/26?
51% chance