The one thing
01GPT-5.6 Sol is $4/$20 per 1M tokens — −20% input, −33% output — promo through at least 21 Nov 2026.
OpenAI’s pricing page and the GPT-5.6 Sol model page both list standard short-context rates at $4 input, $0.40 cached input, $20 output per 1M tokens.12 The model page states that is a 20% cut on input and a 33% cut on output, and that promotional pricing holds at least through 21 November 2026.2 Batch and Flex sit at half those standard rates ($2 / $10).1 Long context (>272K input) doubles input and multiplies output by 1.5 for the full request; cache writes bill at 1.25× uncached input.12 HN carried the pricing link at 316 points / 294 comments inside the window.3
gpt-5.6 still aliases to Sol. Context window 1,050,000; max input 922,000; max output 128,000.2 Kiro now runs the GPT-5.6 family (Sol, Terra, Luna); AWS+OpenAI report ~82% lower cost on successful Terminal-Bench 2.1 tasks for Terra in Kiro — vendor bench, not ours.4
gpt-5.6), re-check invoices and budget alerts against $4/$20. Prefer Batch or Flex for offline work. Pin gpt-5.6-sol if you need the promo rate to stay attached to a snapshot name.Shipped
01MCP Python SDK 2.1.0 · quieter handler errors, leaner content returns
v2.1.0 (24 Aug): Client takes StdioServerParameters directly. Unexpected tool/resource/prompt exceptions now log once server-side and return only Error executing tool <name> to the client — raise ToolError / ResourceError when the model should see the message.5 Tools annotated to return TextContent / Image / Audio (or lists) no longer advertise outputSchema or fill structuredContent; pass structured_output=True to keep the old shape.5 4 MiB body limit extends to SSE and OAuth. v1.29.1 is the maintenance line with the body-limit and recursive-schema fixes.6
Promised, not shipped
OpenAI Private Safety Processing / frontier ZDR — September white paper (issue 013) · Origin agent-native source-control — "ship soon" (issue 011) · Claude Code /design skill — research preview · MAI-Code-1-Flash off Copilot — 10 Sept 2026 · Grok Bot enterprise — waitlist · OpenAI Ultrafast — expands with capacity · Ox Alpha free window — ~through 27 Aug · GPT-5.6 Sol promo pricing — at least through 21 Nov 2026
The conversation
01Sol got cheaper on the API page; MCP Python 2.1 changes what clients see when a handler blows up.15
- OpenAI pricing + GPT-5.6 Sol model pageVendor docs, 25 Aug
- HN · GPT 5.6 Sol price reduction316 pts / 294 cmt, in-window
Community traffic on the pricing URL, not a second primary.3
- X digests · 24 Aug 05:30Digest ~24.5h old — background
Codex quota finish-after-limit reports contested; Claude multi-model errors were open then, resolved since. Colour only.
Update Sol unit economics today. Hold MCP Python 2.1 until you decide whether clients should see handler text or only the generic error line.
GitHub releases: 28 watchlist repos, sweep_complete true, authenticated, 11 in-window (AI SDK monorepo patches collapsed). Vendor feeds: 12 sources, sweep_complete true, 8 in-window. X accounts (~95 roster) and X keyword digests last collected 05:30 UTC 24 Aug (~24.5h old at window close) — background only. Reddit: fresh pulse (6 subs, 13 posts); 24 Aug digest had zero primary posts. HN: Sol pricing thread in-window. arXiv not swept. No issue 22–23 Aug after #015.
Skip this
05Claude Code v2.1.245 (glibc 2.44 startup crash). Arch/CachyOS/Fedora Rawhide only. Upgrade if you hit it; others stay on channel policy from #015.
Anthropic multi-model errors 3.4h major; Cursor Opus 5 errors 3.2h minor; GitHub Fable 5 upstream 0.8h major; Actions start delays 0.6h minor. Resolved weather with duration — not a ship decision.
Kiro + GPT-5.6 / Terminal-Bench 82% cost claim. Vendor bench on Terra in Kiro. Availability note only under One thing.
Cline cli-v3.0.58 / sdk 0.0.79; OpenCode 1.18.22; pydantic-ai 2.34.0; langchain 1.3.17; AI SDK patches. Promo copy, hub log cap, provider guards, catalog rows — no default-stack pin this morning.
Dactyl / Mission Control / Copper / free-token gateways on X. Launches and star counts. No release we verified for a decision this week.
Everything we saw
4848 candidates scanned · 9 used in this issue — the rest, with the reason each one was left out
| Item | Source | Signal | Call |
|---|---|---|---|
| OpenAI API pricing — GPT-5.6 Sol | openai docs | $4/$20 standard; Batch/Flex half | led the issue |
| GPT-5.6 Sol model page | openai docs | −20%/−33%; promo ≥21 Nov | one thing |
| HN — GPT 5.6 Sol price reduction | news.ycombinator.com | 316 pts / 294 cmt | one thing |
| OpenAI — GPT-5.6 in Kiro | openai.com | Kiro GA path; vendor bench | one thing context |
| MCP Python SDK v2.1.0 | github releases | 11.1h | shipped |
| MCP Python SDK v1.29.1 | github releases | maintenance line | shipped note |
| Claude Code v2.1.245 | github releases | 0.9h glibc 2.44 | skipped niche |
| Anthropic status — multi-model errors 3.4h major | anthropic status | resolved major | skipped weather |
| GitHub status — Fable 5 upstream 0.8h major | github status | resolved major | skipped weather |
| Cline cli-v3.0.58 | github releases | promo end + log cap | skipped |
| X digests 24 Aug 05:30 | digests | ~24.5h old | background only |
| Reddit pulse — 6 subs, 13 posts | reddit pulse | layer ran | layer ran |
| pydantic-ai v2.34.0 | github releases | 4.3h | skipped polish |
| opencode v1.18.22 | github releases | 15.5h | skipped |
End of feed. That is everything from the window worth your time.
Next issue tomorrow, 06:00 UTC — and if nothing ships, it will say so in two hundred words.