Claude Haiku 4.5 vs DeepSeek R1 — Side-by-side pricing
Side-by-side API pricing and tokenizer details for Claude Haiku 4.5 (Anthropic) and DeepSeek R1 (DeepSeek).
Side-by-side pricing
| Feature | Claude Haiku 4.5 | DeepSeek R1 |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Input (per 1M tokens) | $1.00 | $0.550 |
| Output (per 1M tokens) | $5.00 | $2.19 |
| Context caching | Yes — 90% off cached tokens | No |
| Batch API discount | 50% off | Not available |
| Context window | 200K tokens | 128K tokens |
| Tokenizer | Anthropic tokenizer | SentencePiece (Llama) |
ⓘ DeepSeek R1: Retired from DeepSeek's own API on 2026-07-24. Figures reflect third-party hosts and are not vendor-published; verify before relying on them.
Real-world cost example
1,000 API requests per month, each with 500 input tokens and 200 output tokens (500K input + 200K output total).
Claude Haiku 4.5
$1.5000
Input: $0.5000 +Output: $1.0000
DeepSeek R1
$0.7130
Input: $0.2750 +Output: $0.4380
DeepSeek R1 is 52% cheaper for this workload — saving $0.7870 per month at this volume.
Frequently asked questions
No, DeepSeek R1 is cheaper for the typical workload above. At $0.550/1M input and $2.19/1M output tokens, it costs $0.7130 versus $1.5000 for Claude Haiku 4.5 — a 52% difference.
Claude Haiku 4.5 supports a 200K token context window. DeepSeek R1 supports a 128K token context window. A larger context window lets you include more text — documents, conversation history, or code — in a single API call.
Claude Haiku 4.5 supports context caching (90% off repeated tokens). It offers a 50% Batch API discount. DeepSeek R1 does not support context caching. It does not offer a batch API discount.
No, they use different tokenizers. Claude Haiku 4.5 uses the Anthropic tokenizer, while DeepSeek R1 uses SentencePiece (Llama). Different tokenizers split text differently, so the same prompt will produce different token counts on each model — the effective cost difference may be larger or smaller than the per-token price difference alone suggests.
Claude Haiku 4.5 (Anthropic): $1 input / $5 output per 1M tokens. DeepSeek R1 (DeepSeek): $0.55 input / $2.19 output per 1M tokens. Rates shown before caching or batch discounts.
For an 80% output / 20% input workload — typical for code generation or long-form writing — Claude Haiku 4.5 costs $4.2000 per 1M total tokens and DeepSeek R1 costs $1.8620. DeepSeek R1 is 56% cheaper for this pattern. For your exact ratio, paste a real prompt into the calculator above.
Claude Haiku 4.5 has a 200K-token context window — approximately 150K words or ~600 pages of standard text. DeepSeek R1 has a 128K-token context window — approximately 96K words or ~384 pages. (Estimates assume ~0.75 words per token.)
Calculate costs for your actual prompt
Paste your prompt into the calculator and get exact token counts using each model's real tokenizer — all in your browser.
Open calculator