ZeroNoise Logo zeronoise
Post
GPT-6.2 caching bug inflates agent bills as Gemini 4 Neon wins practitioners over
•
5 min read
• 147 docs
A Responses API bug stops GPT-6.2 Astra and Sol from caching system prompts; there is a workaround. Also: hands-on Gemini 4 Neon reports, Clay's case for harness work over post-training, and new tooling for governing what agents can do.

Check your GPT-6.2 harness for a caching bug

If you run GPT-6.2 Astra or Sol through the Responses API, the top-level instructions field is not being cached. Codex issue #62312 found it stays at 0 cached tokens on every turn, even though OpenAI's diagnostics wrongly report a cache hit . The reported cost is large. Moving from GPT-6.1 Sol to 6.2 Sol dropped actual cache hit rates from 97.4% to 46.4%, so 64K–72K system tokens were billed at full price on every step .

Workaround: move the system prompt into a leading developer message in input. OpenCode already does this . Peter Steinberger landed a fix in OpenClaw's embedded harness and warns that "most harnesses aren't updated yet" . If you maintain your own harness, look at cached-token counts, not the diagnostics.

Gemini 4 Neon: hands-on reports

Last brief covered Neon's launch. This period brought firsthand usage. Matthew Berman built his open-source AI video editor, Scarlet, with Neon. He says the other frontier models keep needing "explain it more simply" follow-ups, while Neon both writes concise summaries and copes with "highly underspecified prompts" . His examples:

  • Screenshot plus one line. He dropped in a screenshot of the model picker and asked for "make these one line and put the disabled ones below the active ones." About 5 minutes later the change was done. When he asked for a mock, Neon built a live one with toggles, favorites and filters, and said it would include a live mock for UI changes from then on .
  • Filling in his intent. Asked to "add Gemini models to Scarlet like Anthropic and OpenAI are," it spent 45 minutes and routed editing through the Antigravity CLI over agent client protocol, plus the Gemini API for spoken-note cleanup. Berman says that was exactly what he wanted .
  • Computer use. It diagnosed an OBS profile mismatch, fixed the settings, and did not restart recordings that were already running so it wouldn't interrupt them .

Berman says Neon is weaker on 3D demos and games . Riley Brown rates it "very good" for coding and video editing, but its "It's not this — it's this" habit is the worst he has seen, and he advises against using it for writing . That conflicts with Theo's setup in the last brief, where Neon wrote all the copy. Theo now reports "insane results" on long-horizon engineering work but hasn't shared details yet . He says Google is "so far ahead" on making the model easy to work with .

Clay: harness and interface work beat post-training

Clay's Jeff Barg, on LangChain's Max Agency podcast, gave the most reusable engineering lessons of the period:

  • Post-training gave no step change. Their trained model moved closer to Sonnet/Opus, but the team decided "there's a lot of juice" in harness engineering instead .
  • Make your domain look like what models were trained on. Filter-composition structured output handled easy searches but missed long-tail queries. Clay replaced it with a SQL-like query language, which models generate well and which is also composable and modular .
  • Start small. They began with off-the-shelf Pi, chosen because it is light and extensible. They haven't forked it and customize mostly through plugins . The agent has about 20 skills and 8–10 tools, and 85%+ of calls go through bash . The production agent uses the same CLI that external agents get .
  • Debugging long traces. Asking Codex what went wrong in a raw trace with hundreds of tool calls worked "really bad." Having it first summarize each bash command worked much better .
  • Evals: Clay started with capability buckets, then tried coverage buckets per launch milestone, which Barg found "almost meaningless," and settled on user journeys split into subtasks .

Astra and Sol: what they're good at

Berman credits Astra's computer use with fixing frame drops he had chased for 18 months. It read OBS logs during a live recording and forced on a buffer setting that "Auto" had left off . Astra has powered Scarlet's video editing for weeks , although he rechecks Astra-built code with Opus . In a same-prompt comparison he shared from another user:

Model Time Cost
Astra 45 min $9
Sol 30 min $2
Fable 5.2 3 h $200
Opus 5.5 almost 3 h $31

Berman judged Fable's output best, with Astra comparable .

Tools and releases

  • LangSmith Tool Gateway (public beta) is a managed layer that handles credentials, identity and access for every agent tool call . Setup: open Tools, add an MCP server or turn on a built-in, connect your accounts, pick the tools for a kit, and point any MCP client (including Claude Code or Cursor) at the kit's URL . Search mode loads only the tools relevant to the current task . Admins can deny tools or require human approval for sensitive calls .
  • Antigravity desktop can now connect to remote servers over SSH .
  • Solid 2.0 errors include an error code, the exact location, a fix, and a path to a skill.md that tells your agent how to fix it. Some of the last scheduler bug reports came from agents .
  • JetBrains Air (from a sponsor segment) runs Claude Code, Codex, Copilot and other agents side by side inside JetBrains IDEs, with access to tests, inspections and the debugger. Line-level review notes go back to the agent .
  • treg /doomscroll is an open-source skill: give it a topic and a time window, and it finds posts across X, TikTok, YouTube, Reddit, HN and Product Hunt . Repo: superdesigndev/treg .

Security and quality signals

The Pragmatic Engineer covered the Railway hack. An attacker using an abliterated model (one with its safeguards removed) planned a route to AI provider keys within six hours of downloading Railway's monorepo . Railway's fixes:

  • each build gets its own throwaway VM
  • fine-grained access control for CI secrets, with 2FA or biometrics for sensitive credentials
  • security designed on the assumption that attackers already have the source code

Greg Brockman posted "Codex for securing Ruby," quoting a researcher's universal pre-auth Rails RCE. Neither post explains how Codex was used .

Sourcegraph's 2026 Big Code Report says 54% of teams using AI find that more than 60% of AI-generated work needs significant rework, and only 10% say their agents share a common view of the codebase .

GPT-6.2 caching bug inflates agent bills as Gemini 4 Neon wins practitioners over
Summary
Coverage start
1 day ago
Coverage end
14 hours ago
Frequency
Daily
Published
12 hours ago
Reading time
5 min
Research time
4 hrs 38 min
Documents scanned
147
Documents used
17
Citations
37
Sources monitored
110 / 111
Insights
Skipped contexts
Source details
Source Docs Insights Status
LangChain Blog 5 4
Brent Traut 1 1
Lukas Möller 0 0
Jediah Katz 0 0
Aman Karmani 0 0
Jacob Jackson 0 0
Cursor Blog | RSS Feed 0 0
Nicholas Moy 2 0
Mike Krieger 0 0
Sualeh Asif 0 0
Michael Truell 0 0
Google Antigravity 1 1
Aman Sanger 0 0
cat 0 0
Mark Chen 0 0
Greg Brockman 2 1
Tongzhou Wang 0 0
fouad 0 0
Calvin French-Owen 0 0
Hanson Wang 0 0
Ed Bayes 2 1
Alexander Embiricos 0 0
Tibo 8 3
Romain Huet 0 0
DHH 4 0
Jane Street Blog 0 0
Miguel Grinberg's Blog: AI 0 0
xxchan's Blog 0 0
<antirez> 0 0
Brendan Long 0 0
The Pragmatic Engineer 1 1
David Heinemeier Hansson 0 0
Armin Ronacher ⇌ 1 0
Mitchell Hashimoto 0 0
Armin Ronacher's Thoughts and Writings 0 0
Peter Steinberger 0 0
Theo - t3.gg 23 8
Sourcegraph 1 1
Anthropic 2 0
Cursor 0 0
LangChain 1 1
Anthropic 1 1
LangChain 9 3
Cursor 0 0
Riley Brown 0 0
Riley Brown 11 3
Jason Zhou 10 2
Boris Cherny 0 0
Mckay Wrigley 0 0
geoff 4 2
Peter Steinberger 2 1
AI Jason 0 0
Alex Albert 0 0
Latent.Space 0 0
Logan Kilpatrick 5 0
Fireship 0 0
Fireship 1 1
Kent C. Dodds 🐨 13 2
Practical AI 0 0
Practical AI Clips 0 0
Stories by Steve Yegge on Medium 0 0
Kent C. Dodds Blog 0 0
ThePrimeTime 0 0
Theo - t3․gg 1 1
ThePrimeagen 13 2
Ben Tossell 5 0
swyx 12 0
AI For Developers 0 0
Geoffrey Huntley 0 0
Addy Osmani 0 0
Andrej Karpathy 0 0
Simon Willison 0 0
Matthew Berman 2 2
Changelog 0 0
Simon Willison’s Newsletter 0 0
Agentic Coding Newsletter 0 0
Latent Space 1 1
Simon Willison's Weblog 2 1
Elevate 0 0
Lukas Möller 0 0
Jediah Katz 0 0
Sualeh Asif 0 0
Mike Krieger 0 0
Michael Truell 0 0
Cat Wu 0 0
Kevin Hou 0 0
Aman Sanger 0 0
Nicholas Moy 0 0
Andrey Mishchenko 0 0
Jerry Tworek 0 0
Romain Huet 0 0
Thibault Sottiaux 0 0
Alexander Embiricos 0 0
xxchan 0 0
Salvatore Sanfilippo 1 1
Armin Ronacher 0 0
David Heinemeier Hansson (DHH) 0 0
Alex Albert 0 0
Logan Kilpatrick 0 0
Shawn "swyx" Wang 0 0
Jason Zhou 0 0
Riley Brown 0 0
McKay Wrigley 0 0
Boris Cherny 0 0
Ben Tossell 0 0
Geoffrey Huntley 0 0
Peter Steinberger 0 0
Addy Osmani 0 0
Simon Willison 0 0
Andrej Karpathy 0 0
Harrison Chase 0 0