Model comparison
Claude Sonnet 5 vs DeepSeek-V4-Pro-0813
A spec-by-spec comparison of Claude Sonnet 5 and DeepSeek-V4-Pro-0813: pricing, context window, capabilities, and independently sourced benchmarks.
Head-to-head
- DeepSeek-V4-Pro-0813 costs 3.0x less per input token than Claude Sonnet 5 ($0.66/MTok vs $2.00/MTok).
- Claude Sonnet 5 and DeepSeek-V4-Pro-0813 share the same 1,000,000 tokens context window.
- Only Claude Sonnet 5 supports vision.
| Spec | Claude Sonnet 5 | DeepSeek-V4-Pro-0813 |
|---|---|---|
| Provider | anthropic | deepseek |
| Family | Claude Sonnet | DeepSeek V4 |
| Status | frontier | frontier |
| Release date | 2026-06-30 | — |
| Knowledge cutoff | 2026-01 | — |
| Context window | 1,000,000 tokens | 1,000,000 tokens |
| Max output tokens | 128,000 tokens | 384,000 tokens |
| Input price | $2.00/MTok | $0.66/MTok |
| Output price | $10.00/MTok | $1.98/MTok |
| Cached input price | $0.20/MTok | $0.02/MTok |
| Vision | Yes | No |
| Tool use | Yes | Yes |
| Extended thinking | Yes | Yes |
| Streaming | — | — |
Claude Sonnet 5
Strengths- Carries the full 1M-token context window at standard pricing.
- Cheapest 1M-context model from any US lab in this file.
- The scheduled 2026-09-01 increase to $3/$15 was cancelled; $2/$10 is Anthropic’s stated standard price, not a temporary one.
- Uses the newer Claude tokenizer, ~30% more tokens for the same text than Sonnet 4.6 and earlier.
- Anthropic lists classic extended thinking as not supported; only adaptive thinking is available.
DeepSeek-V4-Pro-0813
Strengths- Cache-hit input at $0.022/MTok (off-peak) is roughly 30x cheaper than this model’s own off-peak standard input rate.
- 384K max output is triple the 128K ceiling of the OpenAI and Anthropic flagships.
- Thinking and non-thinking modes are billed at one price.
- Exposes both OpenAI-format and Anthropic-format endpoints.
- Peak-hour pricing (01:00-04:00 and 06:00-10:00 UTC, Monday-Friday, excluding Chinese public holidays) doubles every rate: $1.32/MTok input, $3.96/MTok output, $0.044/MTok cached.
- Vendor label is now DeepSeek-V4-Pro-0813, a dated snapshot; the vendor page previously carried an undated label.
- Does not support image input.
- Concurrency is capped at 500 requests.
- DeepSeek states it reserves the right to adjust prices and recommends checking the page regularly.
Not sure which one to use?
Flaibo is a daily AI orientation hub: a 12-week path and a prompt vault, with Free and Professional plans. Build the habit, not just the comparison.
Start your AI path