Activity
Mon
Wed
Fri
Sun
Dec
Jan
Feb
Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Nov
What is this?
Less
More
21 contributions to Automate What Academy
Claude Fable 5.2
Anthropic dropped a new top-tier model yesterday, and the cost per task is the part I'd look at first. Claude Fable 5.2 is live on the API now, and Anthropic says it beats Opus 5.5 on most of their evals while often costing less per task at lower effort settings. - Model ID claude-fable-5-2, available on the Claude Platform, AWS, Google Cloud and Azure - $8 input and $40 output per million tokens, fast mode is $16 and $80 - Up to 30% faster output than Fable 5.1, per Anthropic - At High effort it roughly matches Opus 5.5 at Max on FrontierCode 2.0 for about half the cost per task - AutomationBench (business workflows): 51.4% vs 42.5% for Opus 5.5, but Gemini 4 Carbon still leads at 53.9% - Mythos 5.2 is the same model with fewer safeguards, only for vetted orgs - Preserved thinking now applies to all API accounts, so check your integrations - Anthropic still calls Opus 5.5 the daily driver and Fable the pick for the hardest work Are you guys going to route your hardest agent steps to Fable 5.2, or is Opus 5.5 still enough for what you're building? For me, Opus 5.5 is plenty for what I need. Read more: https://www.anthropic.com/claude-fable-and-mythos-5-2
1 like • 8d
Agree with you. Sticking with Opus. Might try fable 5.2 just a little on a Saturday in case it burns through my usage so I don’t have to wait long for the weekly reset.
Would you run this from Claude Code or an n8n scenario?
Google just opened up Gemini Notebook to developers, and it ships with an MCP server. The short version: you can now create notebooks and push sources into them from your own code or straight from Claude Code, no clicking around the UI. - REST API plus a remote MCP server, both in Public Preview - 7 MCP tools: create, get, list and update notebooks, plus create, get and list sources - Sources: web pages, YouTube videos, plain text and Markdown - Setup docs for Claude Code and Cursor, and it works with any Streamable HTTP MCP client - Free during the preview, with a limit of 20 new sources per minute per user - Up to 500,000 words per source, 50 sources per notebook on the free tier - Personal Google accounts only for now, Workspace accounts aren't supported yet - Pre-GA, so expect breaking changes The use case I keep coming back to: an agent that watches a client's site, YouTube channel or docs and keeps a research notebook current without anyone touching it. What would you pipe into a notebook automatically if you could? Drop your idea below. Read more: https://developers.google.com/notebook/docs
1 like • 15d
I original thought of using n8n for this, but given how simple the use case you mentioned, I will probably go with Claude Code. Also, my use case is only a spin on yours because I can't come up with a good one atm. Mine is to watch a competitor's site for competitor analysis. Would love to hear what others how others will be using this.
0 likes • 15d
@Ben Williams The client meeting transcripts is an excellent idea, definitely have to wait til they support Google workspace though. I can probably get by creating a free google account that is dedicated to this... so still thinking about it.
Is 51 tokens per second fast enough for your agents?
Another open-weight model just showed up near the top of the leaderboards, and this one is worth a look if you run agents all day. Moonshot shipped Kimi K3.1 Pro on its API, with weights promised by Nov 4, and Artificial Analysis says that would make it the top open-weights model. - Scored 55 on the Artificial Analysis Intelligence Index, 4th of 215 models (independent eval) - $2 per 1M input tokens, $10 per 1M output on Kimi's API - 1M token context window - Slow and chatty: about 51 tokens per second, and 150M output tokens to run the index vs an 87M median - Terminal-Bench 4.0: 65.7% vs Claude Opus 5.5 at 66.4% (Kimi's own numbers) - API IDs: kimi-k3.1-pro and kimi-k3.1-pro-highspeed (Kimi claims 2 to 3x faster) - Weights due by Nov 4, license not stated yet Would you move any of your agent workflows to an open model at this level, or is speed still the dealbreaker? Read more: https://www.kimi.ai/blog/kimi-k3-1-pro
Is 51 tokens per second fast enough for your agents?
1 like • 23d
Too slow, but that's just a guess based on the fact that tool calling is going to make this feel even slower.
Would you trust an agent with your logins this way?
Okay, this one fixes something that has bugged me about browser agents for a while. 1Password just launched a way for AI agents to log into sites for you without ever seeing your actual password. - Agent hits a login page and requests a credential from your vault - You approve it in 1Password, then it autofills the site directly - Agent never sees the password or the one time code - One vault for every agent, so a password change happens once - You can disconnect an integration any time - Free on 1Password Individual and Family plans, not yet on .eu or .ca accounts - Meta's Muse and a Grok agent are rolling out first, Claude, ChatGPT, Perplexity, Browserbase and Browser Use are listed as coming soon Would you hand your agents logins this way, or are you keeping them away from anything that needs a password? I've never used a password manager before (besides Chrome, if that counts). I don't know if I'm about to start now. Read more: https://1password.com/blog/1password-for-ai-agents
1 like • 26d
Your Chrome passwords are stored unencrypted. Look it up and you will start using a password manager very quickly. To answer your question. Yes. I already trust it with API keys (and I moved it out of the .env file). The question is, how many of my passwords do I trust it with? I am not trusting it with my bank accounts yet, but probably soon.
My Thoughts on Opus 5.5
I've been using this for 1 day now. I've noticed that my Claude Code session responses are much quicker. And I like the outputs/explanations slightly more as well. https://www.anthropic.com/claude-opus-5-5
0 likes • Sep 28
Is it still quick 5 days later or are the servers getting slammed now and the quality has degraded? Personally, I think it has degraded, but not down to opus 5 level.
1-10 of 21
Jack CalibratedAI
3
39 points to level up
@jack-calibratedai-6743
AI strategist helping businesses cut through the noise and actually ship. www.caicllc.com

Active 5h ago
Joined Apr 12, 2026
NYC