{"pageUrl":"https://promagen.com/claude-vs-gpt-for-summarising-sources","lastModified":"2026-05-10","provenanceHash":"sha256:0bc153073c4b3eec63a10e3836ec923ca1cfa7a3e3439054fd3926ffb5823d0e","provenanceNote":"Context-window sizes reference vendor-published model documentation. Attribution, citation, and summarisation patterns describe observable model behaviour as of 2026-05-10; model output shifts between releases and is not vendor-guaranteed.","claims":[{"id":"claim-claude-context-window","statement":"Anthropic documents Claude's base context window at 200K tokens, with longer windows available on specific tiers. This is a vendor-documented capability.","evidenceUrl":"https://docs.claude.com","lastVerified":"2026-05-10","hash":"sha256:ef9abfd7a26d6182f741d5bb1c2bf92666e8487e867b105c68b527be2616daba"},{"id":"claim-gpt-context-window","statement":"OpenAI documents the GPT-4-class context window at 128K tokens. This is a vendor-documented capability.","evidenceUrl":"https://platform.openai.com/docs","lastVerified":"2026-05-10","hash":"sha256:fb758ca6ec23de19ea1fb2143f7c4df78206f3771a61776946d318c87e8d49b1"},{"id":"claim-attribution-not-guaranteed","statement":"Neither Anthropic nor OpenAI documents free-form citation/attribution as a guaranteed model output. Attribution discipline depends on invocation mode (retrieval, document context, system prompt) and shifts between model versions; specific attribution-rate numbers are observational, not contractual.","evidenceUrl":"https://promagen.com/sentinel/weekly","lastVerified":"2026-05-10","hash":"sha256:c58591df87d8364ffe8db4f1c3cd1b8fc356261195a10ed7a13ac091a08e26e5"}]}