Opus 5 at half of Sol pricing — same 1M context, different tokenizers.
claude-opus-5 · anthropic
gpt-5.6-sol · openai
Benchmark
| Workload Scenario | Claude Opus 5 | GPT-5.6 Sol | Price Delta |
|---|---|---|---|
| 1M input tokens (raw text) | $5.00 | $5.00 | — |
| 1M output tokens (generation) | $25.00 | $30.00 | −$5.00 |
| 1M tokens · 70% input / 30% output mix | $11.00 | $12.50 | −$1.50 |
| Standard chat turn (4K in / 800 out, 50% cached) | $0.031 | $0.035 | −$0.004 |
| Monthly scale (10K requests / day) | $9,300.00 | $10,500.00 | −$1,200.00 |
Negative difference = Claude Opus 5 is cheaper. Positive = GPT-5.6 Sol is cheaper.
Scaling Curve
Total cost of a token volume at 70/30 input/output split (uncached). Log-log scale.
Verdict
GPT-5.6 Sol has the cheaper input rate, while Claude Opus 5 has the cheaper output rate. The crossover happens when output makes up about 0% of your total tokens.
Below that share (retrieval, summarization, extraction — lots of context in, little text out) GPT-5.6 Sol is cheaper. Above it (generation, translation, coding — long completions) Claude Opus 5 wins.
FAQ
On input, GPT-5.6 Sol is cheaper ($5.00/M vs $5.00/M). On output, Claude Opus 5 is cheaper ($25.00/M vs $30.00/M). For workloads where more than 0% of tokens are output, the output-cheaper model wins overall.
A chat-style request (4,000 input + 800 output tokens, 50% cached) costs $0.031 on Claude Opus 5 and $0.035 on GPT-5.6 Sol — Claude Opus 5 is 1.1× cheaper for that workload.
Claude Opus 5 supports 1,000,000 tokens (128K max output); GPT-5.6 Sol supports 1,050,000 (128K max output). GPT-5.6 Sol fits 1.1× more context, which matters for long documents and agents.