Latency Scaling Differences for GPT and Claude Models | Epoch AI
Epoch AI measures long-context latency scaling across four frontier models. GPT-5.6 Terra and Sol show quadratic time-to-first-token growth with context length, while Claude Sonnet 5 and Opus 5 stay near-linear.
Introduction OpenAI and Anthropic price long prompts very differently. Both have similar context windows: GPT-5.6 models have a maximum of 1.05 million tokens, while Claude models have a maximum of 1 million tokens. However, requests to OpenAI models with more than 272,000 input tokens are charged at double the normal input and cached-input price and 1.5 times the normal output price. Claude models, on the other hand, keep a fixed price regardless of input length. While pricing does not necessarily reflect the underlying inference cost, these pricing differences motivated us to test whether…
saved by
related reading
- Noam Brown on X: "Implications of Large-Scale Test-Time Compute" / Xx.com
- Two different tricks for fast LLM inferenceseangoedecke.com
- AI progress is about to speed up | Epoch AIepoch.ai
- Scaling: The State of Play in AIoneusefulthing.org
- GPT-5.6 Preview System Card - OpenAI Deployment Safety Hubdeploymentsafety.openai.com
- Claude Opus 4.5: Model Card, Alignment and Safetythezvi.substack.com
- News — BenchCADbenchcad.com
- Extending Context is Hard | kaiokendevkaiokendev.github.io
- The Oracle and the Firmcalv.info
- GPT-5.6 Preview System Carddeploymentsafety.openai.com
- How Claude's values vary by model and language \ Anthropicanthropic.com
- Prompting best practicesdocs.anthropic.com