DeepSeek V4-Pro is a mid-tier model priced far below Anthropic's flagship, with a twist: its listed rate is the off-peak price, and it doubles during set weekday hours.
Even at its peak-hour rate, DeepSeek V4-Pro costs a fraction of Claude Opus 5.5 — off-peak it is about 6× cheaper on input and 10× on output. The trade-offs are tier (mid-tier vs flagship), no image input on V4-Pro, and time-of-day pricing you need to schedule around.
| DeepSeek V4-Pro | Claude Opus 5.5 | |
|---|---|---|
| Provider | DeepSeek | Anthropic |
| Tier | Mid-tier | Flagship |
| Input / 1M tokens | $0.66 | $4.00 |
| Output / 1M tokens | $1.98 | $20.00 |
| Cache hit / 1M | $0.022 | $0.20 |
| Cache write / 1M | — | $5.00 |
| Context window | 1M | 1M |
| Notes | A planned Sep 14, 2026 reroute to V4.1-Flash rates was reversed after user pushback; billing unchanged since the Aug 13, 2026 GA build. Off-peak rate shown; peak (01:00-04:00 & 06:00-10:00 UTC weekdays) is 2x: $1.32/$3.96 | New leading Opus model (Sep 22, 2026); Fable 5.1-level performance at 20% less than Opus 5; included on Pro/Max/Team at no extra cost |
DeepSeek V4-Pro is 6.1× cheaper on input tokens ($0.66 vs $4.00 per 1M). DeepSeek V4-Pro is 10× cheaper on output tokens ($1.98 vs $20.00 per 1M).
| Workload | DeepSeek V4-Pro | Claude Opus 5.5 | Difference |
|---|---|---|---|
| Support chatbot 10,000 conversations × (2,000 input + 400 output tokens) | $21.12 | $160 | DeepSeek V4-Pro saves $139 |
| Document Q&A 1,000 questions × (30,000 input + 800 output tokens) | $21.38 | $136 | DeepSeek V4-Pro saves $115 |
| Coding agent session 4M input tokens (90% cache hits) + 150,000 output tokens | $0.64 | $5.32 | DeepSeek V4-Pro saves $4.68 |
Standard rates per 1M tokens from data on this site. The coding-agent row assumes 90% of input is read from the prompt cache (full input price where a model has no cache rate). Cache-write fees, batch discounts and long-context surcharges listed in the notes are not included. Updated November 2, 2026.
you run high-volume text workloads, can schedule batch work outside DeepSeek's peak window, and your evals show a mid-tier model is good enough.
you need flagship-tier output, image inputs, predictable pricing at any hour, or Anthropic's enterprise and data-residency options.
DeepSeek's listed rate is off-peak. From 01:00–04:00 and 06:00–10:00 UTC on weekdays (excluding Chinese public holidays) every rate doubles, cache hits included; weekends are off-peak all day. European morning traffic falls inside the peak window. Even at peak, V4-Pro is well below Opus 5.5's $4.00/$20.00, but your budget should use a blended rate based on when your traffic arrives.
DeepSeek's docs list V4-Pro as text-only (no vision), with thinking mode on by default and up to 384K output tokens. If your pipeline includes screenshots, scanned documents or charts, those calls need another model. Thinking mode generates more output tokens, so measure tokens per task, not just the rate.
V4-Pro's cache hit ($0.022 off-peak) is a thirtieth of its input price, so repeated-context agents are very cheap. Opus 5.5 charges $0.20 per cached read plus $5.00 per 1M tokens to write a 5-minute cache. Anthropic's Batch API halves Opus 5.5 rates for jobs that can wait; DeepSeek's equivalent is scheduling into off-peak hours.
Off-peak, about 6× on input and 10× on output. During DeepSeek's weekday peak window V4-Pro's rates double, so the gap halves.
01:00–04:00 and 06:00–10:00 UTC, Monday to Friday, excluding Chinese public holidays. All other hours, including weekends, are off-peak.
No. DeepSeek's pricing page lists vision as not supported for V4-Pro.
How we choose: prices come from each vendor's official pricing page and are re-checked every week. We take no affiliate fees, so rankings aren't for sale.