Model comparisons, official shutdown and migration guides, and hands-on API notes. Third-party prices and specs are quoted from official sources.
Anthropic opened the Message Threads beta (message-threads-2026-08-12) on 2026-11-02. Checked against the official docs: the 7 ways an API gateway or proxy breaks threads — key rotation 404s, the thread field silently dropped, fixed-field mismatches, cross-key retries, rewritten error codes — and how to fix each on the gateway and the client.
Gemini 4 Argon vs Carbon side by side: what Google has published about each, the same $2/$10 introductory price, and the key difference for developers — Argon is generally available via API, Carbon is in phased rollout. Checked 2026-10-29.
OpenAI removes gpt-4o, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini from the API on January 14, 2027; the official replacement is gpt-6-luna. Full retirement table, the shutdown_date API field, grep patterns, what is not a drop-in, and the complete 2026–2027 OpenAI retirement calendar. Verified 2026-10-26.
Claude Opus 5.5 vs GPT-6.1 Sol side by side: official prices ($4/$20 vs $2/$10), context windows, reasoning controls and tool calling, plus how to A/B both with one byesu key.
Claude Opus 5.5 vs Claude Sonnet 5.5 from Anthropic’s official specs: $4/$20 vs $2/$10, same 1M context and 128K output, different default effort and latency — and when each fits.
GPT-6.1 Sol vs GPT-6 Astra from OpenAI’s official docs: near-Astra performance at one-fifth the list price ($2/$10 vs $10/$50), the same 1.05M context, and when Astra is still worth it.
OpenAI retired gpt-5-codex, the Codex snapshots, chat-latest, and several gpt-4o previews on 2026-07-23. Full model-to-replacement table and migration steps for code that still pins these IDs. Retirement list verified 2026-07-16.
How prompt caching actually bills: a one-time write premium (~1.25x) and cheap cache reads as low as 0.1x. How to structure requests so the expensive prefix is written once and read many times — Claude cache_control breakpoints, OpenAI automatic caching, and how to verify hits. byesu passes caching through unchanged.
Per-image API prices for gpt-image-2, gpt-image-2.5, Nano Banana Pro, Nano Banana 2 and Nano Banana 2.1 on byesu ($0.007–$0.009, same price at 1K/2K/4K), next to OpenAI and Google list prices checked on 2026-10-22.
A fair, source-checked byesu vs OpenRouter comparison: API formats incl. Anthropic Messages, prompt caching billing, fees, payments, image APIs and model coverage.
Claude Opus 4.8 vs GPT-5.6 compared: official list prices, Effort Control vs ultra mode, caching, and workload-based picks — A/B both with one byesu key.
GPT-5.6 Sol/Terra/Luna vs Grok 4.5 compared: official list prices, context windows, output-token efficiency, caching — and how to A/B test both with one key.
OpenAI's Sora API shut down on 2026-09-24. What it affected, how to plan a move to another video provider, and why byesu currently offers image generation only (gpt-image-2, gpt-image-2.5, Gemini image models).
