← Back to all models
D

DeepSeek V2.5 236B

by DeepSeek
Flagship Free & Open Source 🏆 Ranked #39 of 85
76.2
Overall Score
out of 100
About

DeepSeek's 236B MoE model merging V2 Chat and Coder capabilities. A strong open-source model for combined reasoning and coding tasks at manageable inference cost.

Key Metrics
Context Window
128K
tokens
Avg Response
1300
milliseconds
Input Cost
$0.14
per million tokens
Output Cost
$0.28
per million tokens
Arena ELO
1268
Chatbot Arena rating
MT-Bench
8.9
out of 10
Benchmark Scores
MMLU
80.4%
HumanEval
89.0%
MATH
75.7%
GPQA
50.7%
MT-Bench
89.0/10
Capability Profile
Strengths & Limitations
Strengths
✓ Combined chat and coding ✓ MoE efficiency ✓ Open source ✓ Strong reasoning ✓ Cost-efficient
Limitations
⚠ Server-scale hardware required ⚠ Superseded by V3 ⚠ Complex deployment
Ideal Use Cases
Research Coding assistance Complex reasoning Enterprise on-premise Fine-tuning base
Model Details
Provider DeepSeek
Released 2024-09-05
Type Free & Open Source
Multimodal No
Tier Flagship
Global rank #39 / 85
Pricing (USD)
Input tokens $0.14/M
Output tokens $0.28/M
Per 1,000 tokens ≈ $0.0001 input / $0.0003 output
All Benchmarks
MMLU 80.4%
HumanEval 89.0%
MATH 75.7%
GPQA 50.7%
MT-Bench 8.9/10
Arena ELO 1268
Compare this model View Rankings

You might also consider