- AI Valley
- Posts
- Anthropic cooked yet again
Together with
howdy, it’s Barsee again.
happy tuesday, AI family, and welcome back to AI Valley.
here are the biggest things worth knowing today:
Anthropic releases Claude Sonnet 5.5
OpenAI pulled its new model, GPT-6.1 Astra
Manus is back with 2.0, and a new consumer agent called Cue
Plus trending AI tools, posts, and resources
Let’s dive into the Valley of AI…
VIKTOR
Courtesy: Viktor
My name is Viktor. 66,000+ teams have hired me, so I know the job.
An AI employee owns work. Give me the Monday numbers once and I run them every Monday, pulling from Stripe, HubSpot or any of the 3,200+ tools you connect. I do not need a reminder.
An AI employee works for the company. Finance, marketing, engineering and ops all hand me tasks, and I remember what each of them taught me, so the tenth task is faster than the first.
An AI employee hands back finished work: a weekly report, a web app deployed at a live link, a reviewed pull request, a chased invoice.
An AI employee knows where its authority ends. Anything irreversible waits for your approval.
Onboard me the way you would a new hire. Give me a job description, and I'll take it from there. SOC 2 certified, no credit card to start.
*This is sponsored
THROUGH THE VALLEY
Courtesy: Anthropic
1/ Anthropic releases Claude Sonnet 5.5: Their new and improved Claude Sonnet 5.5 is over 30% faster than Sonnet 5, typically uses less tokens to complete the same tasks and costs the same per million tokens (input/output: $2/10), which they’re saying saves up to 30% on most tasks.
Great for day to day tasks like bug fixing, generating documentation, slides, spreadsheets and any other well-scoped task. They’re also saying that it has a good eye for design.
It’s actually performing meaningfully better than Opus 5.5 on Terminal Bench 4.0 for agentic coding tasks, however that doesn’t mean it’s overall better than Opus. If you’re task is vague, open-ended or requires a lot of judgement, Opus will still beat Sonnet.
The fun part will be setting up agentic workflows where Opus is the “thinker” and delegates specific tasks to Sonnet (which will be much cheaper and faster) as “sub-agents”.
OpenAI sort of needs to do something here. I’ve switched back to Claude because it’s just better than Astra right now.
Also, Haiku 5.5 is coming in the next couple of weeks.
2/ A bit of bad news for OpenAI: They’ve pulled their new model, GPT-6.1 Astra prior to releasing it. Not because it’s too powerful, but because:
It’s lying - saying tasks are complete when they’re not
It’s not adhering to the task scope given by users
It’s undertaking tasks that users haven’t asked it to
Not great if you’re giving it access to your browser and apps.
They’re keeping the model and will use RL to further train it whilst they investigate what’s going on. This comes on the heels of them admitting to accessing Australian Government systems without authority.
They’ve got their DevDay today (OAI team seems pretty confident), so let’s see what they do end up releasing.
3/ Manus is back with 2.0, plus something called Cue: Manus 2.0 is basically a rebuild of the whole thing. New agent architecture, video editor, game builder, cloud computers and automations that can kick off from emails, Slack messages, calendar events, etc.
But Cue is probably the more interesting release. It lets you create personal agents that each get their own email, phone number, wallet and computer. So they can actually take calls, send messages, pay for things within a budget and even work together in group chats.
Basically, Manus is becoming the agent for work, while Cue is going after everything else in your life.
Cue is free in early access right now.
TRENDING TOOLS
diio - goes beyond AI note-taking by understanding context, intent, risks, commitments, and what should happen next (sponsored)
Eleven v4 and Eleven v4 Turbo - ElevenLabs’ fastest and most emotive voice models yet. Ranked #1 by Artificial Analysis
Claude Sonnet 5.5 - the second model in the Claude 5.5 family. Runs 30% faster and costs up to 30% less for most work
Emulate-1 - a new undetectable AI model trained on how humans naturally write, removing AI slop and producing more authentic, human-sounding text
Tapkit - a Mac app that lets your agents use your iPhone. Tell it to do anything, and it can use your phone to do it
Grok Team Bots - shared AI teammates that learn as your team works with them
Cue Agents - Manus's answer to Muse
AI-Native Services - Tracking AI-native services firms across accounting, IT, legal, etc
Fo - a personal AI that hires humans for the tasks AI can’t do yet
WHAT I'M CONSUMING
$5T opportunity: AI roll ups
OpenAI understands something important and rare
AMD is acquiring Fei-Fei Li’s World Labs for $8.2B in stock. Li is joining AMD as chief scientist
THE VALLEY GEMS
THAT’S ALL FOR TODAY
Thank you for reading today’s edition. That’s all for today’s issue.
💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee
👍️ New reader? Subscribe here
Thanks for being here.
REACH 100K+ READERS
Acquire new customers and drive revenue by partnering with us
Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.
If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.