Tristan

[Tristan @trssttn]

@theo an enable/disable 5h limit setting for $100/$200 plans would make sense, enabled by default for new users

[Tristan @trssttn]

@steipete hmmm?

[Tristan @trssttn]

French punctuality at its finest. 🙂‍↕️

[Tibo @thsottiaux]

Meant 2pm obviously

[Tristan @trssttn]

@mattpocockuk For folks wanting to try for free: Oracle offers a free 4 OCPUs 24 GB ram 200gb storage arm instance on PAYG accounts or 2/12 on always free accounts (no cc) use your agent to write a provisioning script https://t.co/n2K54X2Bgx

[Tristan @trssttn]

@benkingfm @thekitze yeah I meant more for his workflow, /plan or a workflow that simulates /plan is still very much needed for enterprise work where a mistake can cost lots of $$$

[Tristan @trssttn]

@thekitze let alone an orchestrator supporting multiple providers, models, remote capabilities that is useful, small team, trying to innovate to get ahead, 1k prs 500 issues to triage and more

agree with 1&2

[Tristan @trssttn]

@benkingfm @thekitze he only uses fable/sol afaik so removing the plan makes sense, flagship models don’t really need it, still useful with lower end models or to save the plan/work for future work

[Tristan @trssttn]

also feels like an incredibly easy feature update to market

“Pair once. Claude Code anywhere.”

[Tristan @trssttn]

anyone at ant working on improving `claude remote control` /rc to work more like codex/@t3dotcodes ?

would love a pair-once flow where I can spin up fresh Claude Code sessions from my phone anytime, without needing to manually start /rc for each session @bcherny

[Tristan @trssttn]

@thsottiaux

[Tristan @trssttn]

@thsottiaux update GPT-5.3-Codex-spark to GPT-5.6-Luna
ios notifications when a remote task is done
potential cache miss indicator past 30 min TTL
model/fast emoji/indicator in the chat sidebar (per row)

[Tristan @trssttn]

@ParvezRobinn @theo yes

[Tristan @trssttn]

@HelloGladston @theo yes and a “@t3dotbot ship it”

[Tristan @trssttn]

@polynoamial @OpenAI besides making a few users happy this feels like a ticking time bomb. above 272k input tokens pricing goes up + if the cache is stale past the 30min TTL and the context was well above 272k the weekly usage allowance will drain even faster? https://t.co/MuFQUEMGYu

[Tristan @trssttn]

btw, above 272K input tokens GPT-5.6 Sol is priced at 2x input and 1.5x output for the full request

stick to the default, avoid re-using large sessions after the cache expires, and let their (great) auto-compaction do some magic

[Tristan @trssttn]

From the official GPT-5.6-sol doc https://t.co/FzslyibqyD

Prompts with >272K input tokens are priced at 2x input and 1.5x output for the full request.

Cache writes are billed at 1.25x the uncached input token rate.

[Chubby♨️ @kimmonismus]

Quick reminder: When they first introduced the increased context they said that anything over 272k would consume usage limits at 2x the rate, so be careful when using this!

[Tristan @trssttn]

@jackfriks you should really give the ChatGPT/Codex app a try, imo the remote integration feels way more polished. threads, diffs, screenshots, browser/computer use, etc. all work really well from the phone

just have to ask claude/codex to migrate your CLAUDE.md to AGENTS.md

[Tristan @trssttn]

btw, above 272K input tokens GPT-5.6 Sol is priced at 2x input and 1.5x output for the full request

stick to the default, avoid re-using large sessions after the cache expires, and let their (great) auto-compaction do some magic

[Tristan @trssttn]

One thing to keep in mind with the 1M context:

GPT-5.6 Sol has a 30 min cache TTL (5 min for Fable), so if the cache expires and you come back with 500K+ context, even a small message can become much more expensive.

I think the 258K default makes sense for most users.

[Tibo @thsottiaux]

Here is how to enable a 1M-token context window in Codex for GPT-5.6 Sol.

Even though we have tuned the context limit in Codex to be set optimally when it comes to performance and cost, this is a common ask, so here it is documented.

A larger context window lets Codex retain more code, tool output, and conversation history before summarizing older material. You need a model that supports it. And GPT-5.6 Sol, for example, has a documented 1,050,000-token window.

Open ~/.codex/config.toml and add or update these settings at the top level, before any [section] headers:

```
model = "gpt-5.6-sol"
model_context_window = 1000000
model_auto_compact_token_limit = 900000
```

The first setting selects the model. The second tells Codex to use a one-million-token context budget. The third starts automatic history compaction around 900,000 tokens, leaving some headroom. Restart Codex client and start a new session after saving.

To try the configuration for a single CLI session without changing your defaults:

```
codex -m gpt-5.6-sol \
-c model_context_window=1000000 \
-c model_auto_compact_token_limit=900000
```

Have fun, but also know that we tuned the default carefully!