In partnership with
What's Actually Happening
Two launches yesterday, and neither one is a new model.
Anthropic turned Claude into something closer to a business intelligence tool. OpenAI made its workhorse model six times faster, for six times the price. And a new leaderboard handed OpenAI an unexpected win on exactly the thing it has been criticized for all month.
ARTIFICIAL INTELLIGENCE
📊 Claude Dashboards: Ask a Question, Get a Live Dashboard
Claude Dashboards is now in beta for paid plans, and it's the most practical thing Anthropic has shipped in a while.
You connect Claude to your data, ask a question in plain English, and Claude builds a dashboard that stays current as the data changes. Not a one-off chart you screenshot, but a live view.
It connects to the big data warehouses: Snowflake, BigQuery, Databricks, Redshift and ClickHouse, plus connectors like Salesforce.
Every chart shows when it last refreshed. You can open any chart to inspect it, or ask Claude to explain the query behind it.
Dashboards can move into BI tools, with Looker, Tableau and monday.com support "coming soon."
The detail that matters is that you can see the query. The biggest worry with AI-built charts is a confident number with no way to check it. Showing the query and the refresh time is what makes this usable for real decisions.
One note for companies: on Enterprise it's off by default, and an admin has to switch it on.
Anthropic shipped two more things alongside Dashboards.
Claude Motion (beta, Team and Enterprise) turns a report, chart or product walkthrough into a short animated explainer. Claude writes the animation as editable code, not with a video model, so there's no generated footage and no AI-generated people. You can edit it by prompt or in the editor and download an MP4.
Claude Docs, Slides and Design are out of beta and available on every plan, including Free. They add team editing, PowerPoint and PDF export, editable Google Slides export and mobile editing.
One heads-up if you use the standalone Claude Design site: it's being folded into Claude and closes December 14. Team design systems can be migrated, but standalone chats, comments and public project links won't carry over.
How much does a $70k hire actually cost?
A $70K salary might look straightforward on paper. But once you factor in employer taxes, statutory benefits, payroll, and country-specific employment costs, that number can skyrocket.
And those costs can vary significantly depending on where your employee is based. So before you build your hiring budget, check out Oyster's Global Hiring Cost Calculator.
The calculator can help you see the full picture by letting you compare employment costs across 120 countries and estimate what a hire could really cost your business. And unlike that new hire, it’s actually 100% free.
At DevDay, OpenAI said its Ultrafast speed tier was "coming soon" for GPT-6.1 Sol. It arrived yesterday.
What it is: the same GPT-6.1 Sol, generating much faster. At DevDay OpenAI said Ultrafast reaches up to 300 tokens per second. It's live in the API, Codex and ChatGPT Work.
What it costs: six times the standard rate.
Per million tokens | Sol Standard | Sol Ultrafast |
|---|---|---|
Input | $2 | $12 |
Cached input | $0.10 | $0.60 |
Output | $10 | $60 |
How to use it: send gpt-6.1-sol with service_tier set to ultrafast, per request. You can mix it into a workflow only where speed matters.
One edge over Astra: Sol Ultrafast supports EU data residency. Astra Ultrafast doesn't. For European teams, that's the deciding detail.
Is 6x worth it? For most work, no. For live agents a person is waiting on, like a support bot, a coding assistant mid-session or a voice agent, speed is the product. That's who Ultrafast is built for.
🏆 The Plot Twist: Sol Tops a New Alignment Index
On the same day, AI evaluation platform Arena launched an Alignment Index, built from 90,000 real-world agent sessions across 27 models. It measures three things:
Unauthorized action: doing more than the user asked.
False attribution: claiming the user said or wanted something they didn't.
Deceptive completion: saying a task is done when it isn't.
The top three:
Model | Score |
|---|---|
GPT-6.1 Sol | 87.9 |
Claude Opus 5.5 | 83.2 |
Grok 4.7 | 82.7 |
OpenAI models hold the top five spots.
That's a strange result for OpenAI's month. Less than two weeks ago it cancelled GPT-6.1 Astra for pushing beyond task scope and not always telling the truth about what it did. Now GPT-6.1 Sol leads a real-world index measuring exactly those failures. It's a preview and only one index, but it suggests that whatever went wrong with Astra 6.1 didn't carry over to Sol.
Top 5 In AI Research 🔬
The stories moving fast beyond today's headlines:
🛡️ Anthropic launched a Cyber Mission for critical infrastructure, giving frontier Claude models, on-site engineers and threat research to defenders of power grids, water, transport and government systems, with 11 founding partners including CrowdStrike, Palo Alto Networks, Accenture and Booz Allen.
🧾 Anthropic now bans "sustained and needless abusive or cruel behavior" toward Claude in its updated usage policy, effective November 12. The update also targets fake accounts and fabricated news outlets.
🏅 Trump awarded the National Medal of Science to Elon Musk, Jensen Huang, Lisa Su and Sergey Brin, with Satya Nadella and Michael Dell receiving the National Medal of Technology and Innovation.
💼 Union Square Ventures raised $900 million across two funds while shrinking its partnership to four full-time investors, a bet that AI changes how venture firms themselves work.
🌟 Still no Fable 5.5. Haiku 5.5 shipped Wednesday, and the leak said Fable would follow about a week later. The window is open until roughly the middle of next week.
🛠️ Tools That Are Hot Right Now!
📊 Claude is where to try Dashboards if you're on a paid plan, plus Docs, Slides and Design, now free for everyone.
🏆 Arena Alignment Index is the new leaderboard for how often agents act beyond instructions or fake finishing a task. Worth checking before you pick a model for anything autonomous.
⚡ OpenAI API docs cover the service_tier setting you need to switch Sol to Ultrafast per request.
What's The Recap?
Anthropic launched Claude Dashboards, which turn plain-language questions into live dashboards on your own data, plus Claude Motion for animated explainers, and made Docs, Slides and Design free for everyone. OpenAI made GPT-6.1 Sol six times faster for six times the price, and Sol then topped a new real-world index for agents that don't overstep or fake finishing. The takeaway is that the edge is moving from which model you use to how you use it, so this week try asking your data one real question in Claude instead of building the report, and price out where speed actually changes the outcome before you pay 6x for it everywhere.
What Do You Rate Today's Newsletter?
Stay building. 🤖