Benchmarking Base64 vs Presigned URLs for Multimodal API Image Delivery
Nigiva · reddit · 2026-08-04
When calling multimodal LLM APIs, developers often debate whether to pass images as Base64 Data URLs or presigned S3/R2 links. The author benchmarked both approaches across three vendors to settle the argument with data.
- Team Data URL argues it's operationally simpler: no upload step, no need to regenerate signed links for conversation history.
- Team Presigned counters that keeping completion calls light is better, as resending identical Base64 bytes on every turn bloats the payload.
The writeup details the methodology, traps (like Claude's cachecontrol quirks and accidental cache sharing), and per-model latency results.
More from Infra
- Mach-1 Additive: 35B Model Runs at 120 t/s on Laptops Using 1.7-bit Weights — pbaylies · 2026-08-04
- Surgery on open-weights models: Optimizing inference with hand-rolled Rust implementations — doodlestein · 2026-08-04
- Anthropic Locks in $10B Compute Deal with Volta, a Cloud Startup Just Months Old — The Decoder · 2026-08-04
- Ling-3.0-flash Open-Sourced: 127B MoE with Official FP8 Requiring Only 128GB — derspenti · 2026-08-04
- US ban on Chinese optical transceivers could ironically boost domestic AI hardware, says analyst — bookwormengr · 2026-08-04
- Running DeepSeek V4 Flash on 8x 5090: 50 Concurrent Sessions for $2.5/hr — Hodler-mane · 2026-08-04