CLAUDE.md · price reverts to £5
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Cache Write | Cache Read | Context |
|---|---|---|---|---|---|
| Claude Haiku 4.5 Fast | $0.80 | $4.00 | $1.00 | $0.08 | 200K |
| Claude Sonnet 4.6 Balanced | $3.00 | $15.00 | $3.75 | $0.30 | 200K |
| Claude Opus 4.7 Best | $15.00 | $75.00 | $18.75 | $1.50 | 200K |
| Model | Input (per 1M tokens) | Output (per 1M tokens) | Context | Notes |
|---|---|---|---|---|
| GPT-4o mini Cheapest | $0.15 | $0.60 | 128K | Best for volume |
| GPT-4o Balanced | $2.50 | $10.00 | 128K | Most popular |
| GPT-4 Turbo | $10.00 | $30.00 | 128K | Legacy premium |
| o1 (reasoning) | $15.00 | $60.00 | 128K | Chain-of-thought |
| o3-mini | $1.10 | $4.40 | 128K | Reasoning, budget |
| Use Case | Best Claude Option | Best OpenAI Option | Winner on Price |
|---|---|---|---|
| High-volume simple tasks | Haiku 4.5 ($0.80/M) | GPT-4o mini ($0.15/M) | OpenAI (5x cheaper) |
| Everyday tasks (balanced) | Sonnet 4.6 ($3.00/M) | GPT-4o ($2.50/M) | OpenAI (slightly) |
| Premium quality | Opus 4.7 ($15/M) | GPT-4 Turbo ($10/M) | OpenAI input, Claude output |
| Long docs (200K+ context) | Sonnet 4.6 (200K) | GPT-4o (128K max) | Claude (only option) |
| Repeat system prompt workloads | Sonnet + caching ($0.30/M cached) | No caching available | Claude (10x cheaper when cached) |
| Hard reasoning / math | Opus 4.7 ($15/M) | o1 ($15/M) | Tie |
Claude's prompt caching lets you cache the system prompt and any repeated context. Cache reads cost 90% less than regular input tokens. For applications that send the same large system prompt with every request (chatbots, RAG systems, document Q&A), this can be transformative.
| Scenario | Without Caching | With Caching | Savings |
|---|---|---|---|
| 10K-token system prompt, 1M requests | $30,000 (Sonnet) | $3,000 (Sonnet cached) | 90% |
| 50-page PDF, 10K queries | $1,500 | $165 | 89% |
OpenAI does not offer prompt caching. Google Gemini offers context caching with a 1-hour TTL minimum. Use our calculator to model your workload costs.
| Task | Claude Sonnet 4.6 | GPT-4o | Notes |
|---|---|---|---|
| Summarize a 5-page doc (~2,500 tok) | ~$0.01 | ~$0.008 | Comparable |
| Code review (500 lines, ~3K tok) | ~$0.012 | ~$0.010 | Comparable |
| Process a 50-page PDF (~50K tok) | ~$0.23 | ~$0.19 | GPT-4o cheaper |
| Q&A on entire codebase (200K tok) | ~$0.90 (possible) | ✗ Not possible (128K limit) | Claude only |
| 1M messages/day (1K tok each) | $3,000/day | $2,500/day | GPT-4o cheaper |
For high-volume simple tasks, OpenAI wins — GPT-4o mini is far cheaper than Claude's cheapest option. For workloads with repeated system prompts (RAG, chatbots, document analysis), Claude's prompt caching makes it significantly cheaper in practice. For tasks requiring 200K+ context, Claude is the only option. There's no universal winner — run your own numbers with the Claude Cost Calculator and the prompt pricing tool.
Is Anthropic API cheaper than OpenAI API?
It depends on the tier and use case. At the budget tier, OpenAI GPT-4o mini ($0.15/M) is 5x cheaper than Claude Haiku ($0.80/M). At the mid tier, they're comparable. Claude's prompt caching can flip the equation for certain workloads, making Claude effectively cheaper on repeated prompts.
Does Anthropic offer volume discounts?
Anthropic offers enterprise pricing for large volumes — contact their sales team. The public API pricing applies to most developer use cases. Prompt caching is available on all tiers and is the primary cost-reduction lever.
What is a token in AI pricing?
A token is roughly 4 characters or 0.75 words of text. 1,000 tokens ≈ 750 words ≈ 1.5 pages. Both Anthropic and OpenAI charge per input token (text sent to the model) and per output token (text generated). Use our token pricing tool to estimate costs for your text.
Does Claude have a free API tier?
No. Anthropic's API requires a paid account. There is a free consumer product at claude.ai, but the API has no free tier. OpenAI's API also requires payment. Google Gemini's AI Studio offers a free tier for development and testing.