Drin

I wonder why there's a higher usage of High in @AmpCode team than Everyone. Is it worth using GPT 5.6 high more than medium now that it's more cost efficient? 🤔

@beyang @astrangecat0 one lacking feature in CLI is the ability to see what tags are already attached to the thread and/or search for threads by tags

@astrangecat0 Yep now you can

Amp Orbs make it so easy to fire off threads that now users are asking for better thread management tools. So we shipped thread labels with the ability to add/remove/delete via command palette. Way easier than juggling 5 different agent/app windows/workspaces/tmux sessions.

@beyang can you please add a way to reverse a turn or edit a turn, like in CLI 🙏🏻

@thorstenball i need that Portal tab too 🤯

@dringrayson go to ampcode.com and hit the ship button

Wake up and see in Slack what's been shipped

@sqs can you show percentages on hover? this is very nice!!

We now publish live % usage of Amp mode, for everyone and for just the Amp team. ampcode.com/models

maybe the reason @grok build is such a solid harness is because it uploads everything to their servers 😹

Another @AmpCode proposal: add a way to ensure it remembers whether the sidebar is closed on a given thread or provide a way to set defaults. I like it not being open unless I open it, even after a reload.

Proposal: Update @AmpCode low mode executor to use Grok 4.5 instead of GLM 5.2 🙏🏻

/goal do my job. periodically check Slack and respond to any messages. check JIRA for new tickets assigned to me and if you see one, spawn a subagent to complete the task and open a PR. if you need clarification, reach out to people on Slack for help. NEVER tell them that you’re GPT. make no mistakes.

goddamn grok 4.5 is so good

@bedesqui @sqs brb copying this from ur setup 🤣

@sqs OMFG QUINN I SHIT YOU NOT i was squashing the agent “recipes” i have on my pi(amp knockoff setup, of course) into a single dial the agent can use when spawning sub-agents this afternoon

Intelligence is generalizing, so The Dial makes more sense than smart/deep/rush. Also for the first time, an open model powers one of Amp's builtin agent modes. It is good and cheap! (Workspace admins can disable this, but hey, soon I think we'll have a USA-trained option here.) Also, we now expose a raw model selector if you want something lower-level or to play around with a model before we've found a fit for it in The Dial.

@sqs @JebsSteve0x1 isn't it fable?

@JebsSteve0x1 Opus is the main model for Ultra and the oracle for high

@nicolaygerold can we have a longer system prompt 🤣 Enjoying it thus far

Send me your feedback, tune it with me ;)

@ryan0x44 how did you enable spawning in different dirs?

Ok not sure how this is going to go but we'll see

AHAHAHHA Anthropic will have to offer Fable permanently and drop the prices to stay relevant looool

I fucking love @AmpCode’s newest change to their model selector. It’s one I’m trying to envision how it should be. Frontier models working together and easy enough to change or update depending on your needs and tasks.

@bridgemindai wth are u building using 4 fable accounts?

TODAY IS THE DAY. GPT 5.6 launches. I expect it to crush Fable 5. Already #1 on TerminalBench at 91.9%. Fable 5: 84.3%. Up to 750 tokens per second on Cerebras. And OpenAI's subscription limits have never forced anyone to buy 4 accounts. I am paying $800 a month across 4 Claude Max subscriptions for ONE usable week of Fable. That era ends if GPT 5.6 delivers. One open question: OpenAI shipped its heaviest safety stack ever with this model. Today we find out if its guardrails veto real work like Fable 5 does. GPT 5.6 goes into BridgeBench the second it drops.

fuck and now there's Muse Spark 1.1!!!

Grok 4.5 solved a task for me that only GPT 5.5 xhigh also completed. Grok solved it in less time as well. Very impressed with this model.