The Ledger 13 tracked entriesnofeed.dev
No Feed

No feed. One issue a day. Then you are done.

Weekdays — what shipped, what to skip, what it means. Two minutes on a quiet day, four on a heavy one.
Saturday — one argument, made properly. Sunday — nothing at all.

Every claim links to a primary source. Community posts choose what we look at; they never carry a fact on their own. Corrections run above the fold.

The Ledger

Prices, limits, context windows, parameter changes and shutdown dates, with the date each one was read. Nobody maintains this systematically, which is exactly why it is here — and it is a by-product of the watchers that feed the daily issue, not extra work.

Shutdown clock

anything under fourteen days also appears in the issue ticker
SubjectDateDays left
claude-sonnet-4-5-202509292026-09-2954d
claude-haiku-4-5-202510012026-10-1570d
claude-opus-4-5-202511012026-11-24110d

Everything tracked

newest change first
SubjectKindWasNowEffective
claude-opus-4-5-20251101ShutdownNot sooner than 2026-11-242026-11-24
claude-haiku-4-5-20251001ShutdownNot sooner than 2026-10-152026-10-15
claude-sonnet-4-5-20250929ShutdownNot sooner than 2026-09-292026-09-29
claude-opus-4-1-20250805
Deprecated 5 June 2026. Confirmed still listed Deprecated with tentative retirement August 5, 2026. Requests after the retirement date fail.
ShutdownRetires 2026-08-052026-08-05
Qwen3.8-Max — context window
2.4T MoE total parameters, 95B active per Alibaba press release.
Context1M tokens2026-08-03
Qwen3.8-Max — input / output per 1M
API generally available 3 Aug 2026 via Alibaba Cloud Model Studio. Open weights promised next week; license not yet disclosed. Alibaba press release does not list token rates; $2/$6 read from qwencloud model page pricing strip and corroborated by multiple reseller pages.
Price$2.00 / $6.002026-08-03
temperature, top_p, top_k on Claude Opus 4.7+
Omit the parameter and steer with prompting instead.
ParameterAcceptedReturns 400 on a non-default value2026-08-01
deepseek-v4-flash — context windowContext1M tokens2026-07-31
deepseek-v4-flash — input / output per 1MPrice$0.14 / $0.282026-07-31
GPT-5.6 Sol — Fast mode
Requests tagged priority continue to work.
PricePriority Processing2× price, up to 2.5× faster2026-07-30
GPT-5.6 Terra — input / output per 1MPrice−20% from prior$2.00 / $12.002026-07-30
GPT-5.6 Luna — input / output per 1M
Announced as an 80% reduction. Prior figure inferred from that percentage, not published directly.
Price$1.00 / $6.00$0.20 / $1.202026-07-30
deepseek-chat, deepseek-reasoner
Legacy aliases. Verify what your provider resolves them to before assuming they still work.
ShutdownDiscontinued 2026-07-242026-07-24

Where a figure is inferred rather than published — a prior price reconstructed from an announced percentage, for instance — the note says so. An inferred number is not a quoted one.

Changes land in the morning issue.

Or Saturdays only, if the week is enough. Full issue in the email — no teaser, no click required.