DeepSeek-V3 (MoE) — pricing & specs
DeepSeek-V3 (MoE) is a balanced model from DeepSeek, at capability tier 5 of 8. Live $/M-token pricing and specs from the Mintok catalog.
$/M-token pricing
Open-weights model — no per-token list price. Size your self-host $/M-token →
Specifications
- Vendor
- DeepSeek
- Family
- DeepSeek-V3
- Capability tier
- 5 / 8
- Size class
- BALANCED
- Category
- GENERAL
- Parameters
- 671B total · 8B active (MoE)
- Context window
- 128,000 tokens
- Licensing
- Proprietary
Carbon footprint (modeled)
Modeled decode on a reference NVIDIA Rubin Ultra VR300 at 50% MBU, batch 32, global-avg grid. Compare all models & grids →
Alternatives at tier 5
Same capability tier, sorted by output price — the cheaper-capable options.
- DeepSeek V3DeepSeek$0.80/M out
- Mistral Large 2Mistral$1.50/M out
- DeepSeek R1DeepSeek$2.50/M out
- Gemini 2.5 FlashGoogle$2.50/M out
- Grok 4.3xAI$2.50/M out
- o3OpenAI$8/M out
FAQ
How much does DeepSeek-V3 (MoE) cost per token?
DeepSeek-V3 (MoE) is distributed as open weights — there is no per-token list price; you self-host it. Use the Mintok sizing report to estimate your $/M-token at your volume.
What is DeepSeek-V3 (MoE)'s context window?
DeepSeek-V3 (MoE) supports a 128,000-token context window.
What capability tier is DeepSeek-V3 (MoE)?
DeepSeek-V3 (MoE) sits at capability tier 5 of 8 on the Mintok scale (BALANCED size class).
Is DeepSeek-V3 (MoE) open-source or proprietary?
DeepSeek-V3 (MoE) is a proprietary model from DeepSeek, accessed via API.
Put DeepSeek-V3 (MoE) to work
Benchmark it on your prompts, or size a deployment and compare $/M-token against alternatives.
Pricing and specs are maintained by Mintok's catalog watchers from public sources. Not investment advice. Methodology.