The one thing

GPT-6.1 Sol is out at Sonnet 5.5's per-token price, with cached input at half Anthropic's rate. For agent loops that re-read context, it goes on your eval this week.

OpenAI released gpt-6.1-sol at DevDay on 29 September. API list price is $2 per million input tokens, $0.10 cached input and $10 output.¹ It is live in Codex and ChatGPT Work for Plus, Pro, Business, Enterprise and Edu, and not yet in ChatGPT's Chat.¹ GitHub made it generally available in Copilot the same day for Pro+, Max, Business and Enterprise, billed at provider list price, with a gradual rollout.²

Claude Sonnet 5.5, released a day earlier, lists at the same $2 and $10, with cache reads at $0.20.³ On per-token price the two are identical except for cached input, where Sol costs half. Agent sessions are dominated by cached reads, so that line can decide the bill; how much depends on your cache-hit rate, which only your own logs can tell you.

VERDICT · USE IT — Run it against your own task set before you move traffic. The price is verified; the capability claims are OpenAI's own runs.

What OpenAI claims, and what else DevDay shipped

OpenAI says Sol matches GPT-6 Astra on DeepSWE v1.1 at roughly a fifth of the cost, and scores 2.2 points above Opus 5.5 on AutomationBench at medium effort.¹ Both are vendor measurements and we have not seen them reproduced.

Also live from the same event: GPT-6 Astra Ultrafast in the API and in Codex on Pro 500 and Enterprise; Codex in the cloud with reusable environments; the Agents API with computer use.⁴

Shipped

Use it

pydantic-ai 2.52.0 and 1.107.7 patch a denial-of-service in `web_fetch`

Advisory GHSA-v36g-jcw9-x7cw, rated moderate: converting attacker-controlled HTML with deeply nested elements in the local web_fetch tool could consume excessive CPU and memory. Provider-native web fetching is not affected.⁵ Both release lines are patched.

MATTERS TO · Anyone whose pydantic-ai agents fetch pages they do not control. Upgrade either line today.

nofeed.dev/issues/2026-09-30/pydantic-ai-2-52-web-fetch-advisory/

Breaking

Codex CLI 0.159.0 removes prompt suggestions and the bundled `plugin-creator` skill

The release drops automatic follow-up prompt suggestions along with the tui.prompt_suggestions setting, and removes the bundled plugin-creator skill.⁶ Patch releases 0.159.1 and 0.159.2 followed inside the window.

MATTERS TO · Teams whose scripts or onboarding docs call `plugin-creator`, or who pin Codex config files. Check before upgrading.

nofeed.dev/issues/2026-09-30/codex-cli-0-159-removals/

Promised, not shipped
[object Object] · [object Object] · [object Object]

The conversation

The claim — including ours

The Sol thread on Hacker News (867 points, 786 comments) argued less about Sol than about GPT-6 Sol, released days earlier.

From the floor

the_duke, Hacker Newsreader-posted, unverified
Called GPT-6 Sol a regression from Sol 5.6 and said they had switched to Opus 5.5 for coding.
proxysna, Hacker Newsreader-posted, unverified
Said they have not spent $200 on DeepSeek this year and find the intelligence gap negligible.

Our take

A 6.1 arriving this soon after 6.0 fits the regression reports, but nothing published confirms why. Treat any Sol result from before 29 September as a result about a different model.

Hacker News swept in full: 38 stories in the window, 10 above 80 points, read with comments. Vendor changelogs and status pages: 12 feeds, 16 entries. GitHub releases across 51 watched repositories: 33 in the window. arXiv, 48 hours: 400 fetched, 165 kept, top 12 read. X did not run for this window: the newest keyword digest and roster digest (100 accounts) were both collected 29 September 05:30 UTC, about 24.5 hours old and before DevDay, so background only. Reddit: the newest digest is from the same hour, and the fallback pulse across six subreddits returned no posts.

In production

59 minutes and 5 h 22 min

Claude.ai, Claude Code, Cowork and the API returned errors from 14:00 to 14:59 UTC on 29 September; sign-in and new Claude Code sessions were among the last to recover, and some messages sent in that hour may not have been saved.⁷ OpenAI's API, Codex and ChatGPT ran degraded from 17:52 to 23:14 UTC the same day, with an RCA promised within five business days.⁸Source · Anthropic and OpenAI status pages

Durations are the impact windows each vendor published, not the time the incident page stayed open.

Skip this

Sol matching Astra on DeepSWE, beating Opus on AutomationBench OpenAI's runs of OpenAI's model. We print their prices and leave the benchmarks out.

Dots, OpenAI's always-on agents Rolling out inside ChatGPT Pro and Business Premium. No API, nothing a developer builds on yet.

ChatGPT Pro 500 Named in the DevDay recap; the help page we read lists only the $100 and $200 tiers and no usage figures.

Anthropic's GLM-5.3 cyber-capabilities post Policy research. Nothing to install, pin or stop using.

Claude Code 2.1.285, Gemini CLI 0.62.0, LiteLLM 1.103.1 No breaking, deprecation or removal sections. Upgrade on your usual schedule.

Everything we saw

71 candidates scanned · 7 used in this issue — the rest, with the reason each one was left out

The receipt. Ranking is only trustworthy if the discarded pile is visible, so here it is: everything the collectors surfaced in the window, with its signal and what we did with it.

Item	Source	Signal	Call
Introducing GPT-6.1 Sol (HN 867 points)	OpenAI News	New model, API price published	Led the issue
GPT-6.1 Sol in GitHub Copilot	GitHub Changelog	GA, gradual rollout	Used in the lead
DevDay 2026 Recap (HN 91 points)	OpenAI News	20+ announcements	Used for Promised and the lead's deeper block
pydantic/pydantic-ai 2.52.0, 1.107.7	GitHub Releases	Security advisory, moderate	Printed as Use it
openai/codex 0.159.0, 0.159.1, 0.159.2	GitHub Releases	Setting and bundled skill removed	Printed as Breaking
Claude: elevated errors across claude.ai, Claude Code, API (HN 171 points)	Anthropic Status	Major impact, 14:00 to 14:59 UTC	In production
OpenAI: elevated errors across ChatGPT, Codex and the API	OpenAI Status	Degraded, 17:52 to 23:14 UTC	In production
Introducing dots (HN 534 points)	OpenAI News	Consumer agent, Pro and Business Premium	Skipped
About ChatGPT Pro tiers (HN 197 points)	OpenAI Help	No Pro 500 figures on the page	Skipped
GLM-5.3 and the spread of advanced cyber capabilities (HN 211 points)	Hacker News	Research post	Skipped
Cursor status: xAI models degraded; Cloud Agents and Grok Bot degraded	Cursor Status	Major for 0.6 hours; minor for 3.7 hours	Resolved inside the window
OpenAI status: ChatGPT Pro/Plus errors, Space Pages errors, support chat down	OpenAI Status	Minor or none; two open at cut-off	ChatGPT-only, weather
GitHub: Dependabot custom runner settings; external custom properties	GitHub Changelog	Admin features	Off the beat
Vercel: Connect service submissions; trace span search from the CLI	Vercel Changelog	Platform features	No decision changes
Hugging Face: NVIDIA Kumo Tabular; source-aware verification for MCP agents	Hugging Face Blog	Vendor posts	No independent evaluation
anthropics/claude-code 2.1.285; google-gemini/gemini-cli 0.62.0; cline 4.1.22, CLI 3.0.66, Desktop 0.0.38	GitHub Releases	No breaking sections	No decision changes
openai/openai-python 3.21.0, 3.22.0, 3.22.1; vercel/ai 7.0.123, 6.0.297, 5.0.270; langchain-core 1.6.6, langchain-anthropic 1.7.5	GitHub Releases	Feature and patch releases	No decision changes
BerriAI/litellm 1.103.1, 1.102.2, 1.101.3	GitHub Releases	Patch releases across three lines	No decision changes
mvschwarz/openrig 0.6.1, 0.6.2; pbakaus/impeccable engine 0.1.8; Tencent/BrowserSkill 0.3.2; BuilderIO/agent-native; stablyai/orca 1.4.217; alibaba/open-code-review 1.12.11; akitaonrails/ai-memory 2.4.2; vectorize-io/hindsight 0.10.2	GitHub Releases	Patch cadence; openrig and impeccable newly admitted to the watchlist today	No decision changes
arXiv 2609.37864: AgentBug-Smith, reproducing agent-harness bugs	arXiv	Benchmark construction for harness bugs	Abstract only; watch
arXiv 2609.36570: CounterSteer, activation steering against indirect prompt injection	arXiv	Inference-time defence, per-model recipe	Abstract only; needs open weights to apply
arXiv: 10 further agent and SWE papers (harness optimisation, context compression, SWE verification)	arXiv	Benchmark gains	No change to practice yet
DraftKings AI targeting; AI $6T revenue estimate; Meta Muse permissions; privacy analysis of AI chat apps	Hacker News	154 to 537 points	Off the beat
X digests (keyword, roster) and Reddit digest	Community digests	Newest files 24.5 hours old, collected before DevDay	Background only; Reddit fallback returned nothing

EOF

End of feed. That is everything from the window worth your time.
Explore more news, analysis and practical guides on NO FEED.

Still developingClaude Sonnet 5.5's five migration breaks, from issue 41, still apply to anyone switching models this week.

Sources — each with the date it was read01 OpenAI: Introducing GPT-6.1 Sol (2026-09-30) · 02 GitHub Changelog: GPT-6.1 Sol in GitHub Copilot (2026-09-30) · 03 Claude Platform release notes, 28 September 2026 (2026-09-29) · 04 OpenAI: DevDay 2026 Recap (2026-09-30) · 05 pydantic-ai 2.52.0 release notes (2026-09-30) · 06 Codex 0.159.0 release notes (2026-09-30) · 07 Claude Status: Elevated errors on claude.ai, Claude Code, Claude Cowork and the Claude API (2026-09-30) · 08 OpenAI Status: Elevated errors across ChatGPT, Codex, and the API (2026-09-30)