DeepSeek V4 Pro
Flagship open-source MoE (1.6T) — frontier reasoning + coding with a 1M-token context, at open-weights pricing.
Capabilities
Pricing
| Input | $1.74 / 1M tokens |
| Cached input | $0.145 / 1M tokens |
| Output | $3.48 / 1M tokens |
Provider-equivalent rates used for credits. Higher reasoning effort can consume more output tokens; it has no separate fixed multiplier. One credit equals 1¢, with a one-credit minimum per completed run and the aggregate rounded up once.
Provider pricing source · verified 2026-07-20
Compare DeepSeek V4 Pro
DeepSeek V4 Pro: frequently asked questions
- What is DeepSeek V4 Pro?
- Flagship open-source MoE (1.6T) — frontier reasoning + coding with a 1M-token context, at open-weights pricing. You can try DeepSeek V4 Pro on TryAI without a separate provider account.
- How much does DeepSeek V4 Pro cost on TryAI?
- DeepSeek V4 Pro uses these provider-equivalent rates: Input: $1.74 / 1M tokens, Cached input: $0.145 / 1M tokens, Output: $3.48 / 1M tokens. One TryAI credit equals one cent of usage; completed runs have a one-credit minimum and the aggregate cost is rounded up once.
- What can DeepSeek V4 Pro do?
- DeepSeek V4 Pro supports Streaming, Reasoning, Tool use with a context window of about 1M tokens.
- How do I try DeepSeek V4 Pro?
- Create a free TryAI account, pick DeepSeek V4 Pro from the model selector, and start generating in seconds.
Related models
Claude Opus 5
Chat & reasoning
Anthropic's everyday flagship for complex agentic coding and enterprise work. Ranks first overall on the AA Intelligence Index v4.1 at 63 with max effort.
Try Claude Opus 5→Claude Fable 5
Chat & reasoning
Anthropic's highest-capability model for the hardest reasoning and long-horizon agentic work. Scores 62 on the AA Intelligence Index v4.1, just behind Opus 5 at max effort.
Try Claude Fable 5→Grok 4.6
NewChat & reasoning
xAI's frontier model for long-running agents, codebase work, and ambitious interactive or visual projects. Scores 61 on the AA Intelligence Index with 500K context and always-on reasoning.
Try Grok 4.6→