DeepSeek-V3: Benchmarks, Pricing & Context Window
DeepSeek-V3 is a language model from DeepSeek, released in December 2024, with a 164K-token context window, and pricing from $0.320/M input and $0.890/M output.
A powerful Mixture-of-Experts (MoE) language model with 671B total parameters (37B activated per token). Features Multi-head Latent Attention (MLA), auxiliary-loss-free load balancing, and multi-token prediction training. Pre-trained on
DeepSeek-V3 benchmarks
Capability tiers
Standing within each category, adjusted for leaderboard depth.
Real tasks performance
High-confidence performance for DeepSeek-V3 across real-world prompt categories. Only 95% intervals at most 4 points wide are shown.
Performance by conversation depth
How DeepSeek-V3 holds up as conversations get longer.
Quality Tracker
DeepSeek-V3 Performance Across Datasets
Scores sourced from the model's scorecard, paper, or official blog posts
llm-stats.com - Thu Oct 08 2026DeepSeek-V3 pricing
Providers
DeepSeek-V3 starts at $0.270 per million input tokens and $1.10 per million output tokens via DeepSeek. See all 2 providers below with their per-token pricing, latency, throughput, and modality support.
| Provider | Input $/M | Cached input $/M | Output $/M | Context in / out | TTFT p95 s | Output p5 c/s | Modalities in / out |
|---|---|---|---|---|---|---|---|
 DeepSeek | $0.270 | — | $1.10 | 131.1K/131.1K | 0.50 | — | / |
 DeepInfra | $0.320 | — | $0.890 | 163.8K/163.8K | — | — | / |
Cached input is the discounted price for prompt tokens served from a provider cache. TTFT is time to first token. Output is characters per second; p5 is the sustained floor exceeded by 95% of observed requests.
DeepSeek-V3 model size
DeepSeek-V3 has 671 billion parameters and was trained on 14.8 trillion tokens. See how it compares to other models in the same parameter range.
DeepSeek-V3 context window
Input and output token limits for DeepSeek-V3, plus how it ranks on long-context understanding.
Try now
huggleMake it with
DeepSeek-V3.
DeepSeek-V3 latency
DeepSeek-V3 time to first token, sustained output throughput, and failed-request rate from live model usage over the trailing 7 days.
DeepSeek-V3 examples
Recent arena outputs from DeepSeek-V3, picked from the highest-ranked matchups.
DeepSeek-V3 license
DeepSeek-V3 is released under the MIT + Model License (Commercial use allowed) license, which restricts commercial use, has 671.0B parameters.
- License
- MIT + Model License (Commercial use allowed)
- Non-commercial
- Parameters
- 671.0B
DeepSeek-V3 resources
Official sources for DeepSeek-V3: provider documentation, official playground, paper or system card, source repository, model weights.
DeepSeek-V3 vs other models
The most-compared alternatives to DeepSeek-V3 are Claude 3.5 Sonnet, Phi 4 Reasoning Plus, GPT-4o. Open any pair side-by-side for benchmarks, pricing, context, and latency.
Models like DeepSeek-V3
Models ranked just above and below DeepSeek-V3 by LLM Stats score.
FAQ
Common questions about DeepSeek-V3.