😺 Claude is a financial analyst now

PLUS: AI improves at human tasks how often?!

Welcome, humans.

The rumor on X is that ChatGPT-5 will be released around July 31… but of course, take that with a grain of salt. As you might remember, late last week (or was it early this week? What is time??) Sam Altman announced they would indefinitely delay the open source reasoning model they were planning in order to further red team it.

That model was SUPPOSED to come out this Thursday. We’re now not sure which will come first; GPT-5 or “open o1” (not the official title)…

After the success of Grok 4, perhaps they are taking their time to get these releases right… and if the leaked benchmarks on X are to be believed (again, salt grains ppl), it might be juuuust ahead of Grok 4 Heavy in terms of intelligence.

Anyway, we share all this to say we might be looking at a slow rest of the month for OpenAI’s release calendar. You’re of course welcome to prove us wrong, OpenAI…

Here’s what you need to know about AI today:

Claude released a new finance tool for financial service providers.
Mira Murati's Thinking Machines raised $2B at $12B valuation with no products.
Mistral released a new AI transcription tool that beats OpenAI’s.
60% of US managers use AI for personnel decisions, 78% for raises.

Advertise in The Neuron here.

P.S: If you use Gmail and this email gets cut off, click: “[Message clipped] View entire message” so you don't miss anything!

AI's got a new job: replacing your Bloomberg Terminal.

Just when you thought AI was running out of industries to disrupt, Anthropic just released Claude for Financial Services… and suddenly, that $25K-a-year Bloomberg Terminal is looking a bit... vintage.

Here's what makes it different:

Claude can now combine external market data (from S&P/FactSet/Monginstar) with the firm's own internal data (Databricks/Snowflake) to answer questions like “How did our tech portfolio perform vs. the S&P 500 tech sector?” without analysts having to manually pull from multiple systems
Need to run Monte Carlo simulations? Build proprietary trading models? Claude Code handles the heavy lifting, with expanded usage limits for those all-nighters before earnings calls.
No more jerry-rigging APIs on your own. Claude comes with pre-built MCP connectors ready to plug into your existing financial data infrastructure.
Expanded limits that bump analysts from regular Claude's 200K token limit (about 500 pages) to higher capacity thresholds and more queries per hour, so they can process massive 10-K filings and deal docs without hitting walls during market deadlines.

Oh, and did we mention big names are already believers?

Bridgewater estimates 20% productivity gains.
Norway's sovereign wealth fund (NBIM) saved 213K hours.
AIG compressed underwriting review time by 5x.

Here’s one reason why: On the latest finance agent benchmark, Claude Sonnet 4 hit 44.5% accuracy on complex financial analyst tasks; neck and neck with OpenAI's o3. For context, most models scored below 30%.

These aren't simple “What's Apple's stock price?” questions – we're talking multi-step SEC filing analysis that would make a first-year analyst sweat.

Claude isn't alone in this financial AI arms race. Perplexity is another key player in the AI finance race, targeting individual analysts with three products:

Perplexity Finance ($0): Free real-time stock prices, company analysis, and 13F comparisons…all wrapped in what they call a “delightful UI.”
Enterprise for Financial Services ($40/month): FactSet integration and industry research. Teams save 10+ hours weekly per employee. One analyst said it summarized 48 hours of Q4 earnings work in 2 minutes.
Perplexity Labs ($20/month): Custom dashboards and deep research reports in 10+ minutes.

The best story? Stanley Druckenmiller used Perplexity to identify the top five Argentine ADRs, bought them all, and his positions have grown significantly

And then there’s the crypto connection: Last week, Perplexity partnered with Coinbase to integrate real-time crypto data.

Phase one launched with COIN50 index prices in Perplexity's Comet browser.
Phase two connects queries directly to Coinbase trading.
CEO Brian Armstrong's vision: crypto wallets fully integrated into AI models.
He believes this will create “another 10x unlock” for AI.

Imagine asking your AI to not just analyze Bitcoin, but actually buy it for you. Cue the obligatory “what could go wrong??” but I mean, yeah, who has the time to keep up with crypto trading, that stuff is going on 24 hours a day.

Why this matters: ~70% of US stock trading already runs through AI algorithms. This isn't about replacing humans—it's about upgrading their tools. And both tools have potential benefits:

Claude's edge: Enterprise security, institutional data integrations, fantastic coding tools to build your own tech on top, and serious financial backing.
Perplexity's edge: Accessibility, customizable dashboards, and better user experience. At $20-40/month vs. Bloomberg's ~$25K/year, it targets a much broader market… but financial pros are using it too.

Our take: This is natural evolution. The hodgepodge mess of financial data is finally getting the AI upgrade it desperately needed. Financial firms have used AI algorithms to trade for years. Now they're using AI to understand what to trade. The winners will be anyone who's struggled through a 10-K at 2 AM… or eventually wants to fire their financial advisor.

Think about that last part: when we get into the realm of agentic trading accessible for everyone, it won’t just be quant firms with a leg up to trade with; you’ll be able to tell your AI “I want to invest in these assets, hold these assets, and ditch these assets; maximize profits, minimize losses, and warn me if I’m making a mistake.”

While that type of 24/7 financial advisor on demand is still years away for us normies, the near term impact of these tools could be The Bloomberg Terminal becoming Bloomberg toast—not because these tools are perfect, but because they're solving the real problem: making financial data actually usable.

FROM OUR PARTNERS

Crush your next project with Clutch!

You’ve got a big project to launch—and no time to waste vetting endless options. That’s where Clutch comes in. Just tell them what you need, and they’ll instantly match you with verified dev partners based on your goals, timeline, and budget.

Powered by AI, informed by thousands of real reviews, and built to remove the friction from finding the right team. Whether you’re building an app or a custom platform, Clutch helps you move fast and move smart.

Share your project goals with Clutch

Prompt Tip of the Day

Today we’re going to cover two simple requests you can include in your prompts to encourage your thinking AI models to think hard.

For starters, try adding “ultra think” and “think ultra hard” in your prompts (this is advice from Kieran Klaassen, via Peter Yang).

The AI creators have suggested that using terms like “give it all you got” and “think hard” help encourage the model to really think on something (they also have “thinking budgets” you can edit in the API).

The second idea is to use “really think. Really, really think.” That’s from Matt Shumer, who shared this prompt on X that helps for UX/product design.

Try them both, see if it helps!

Check out all of our Prompt Tips of the Day from June here.

P.S: Completely new to AI? Start here!

Treats To Try.

*Asterisk = from our partners. Advertise in The Neuron here.

*Murf AI turns your text into human-like speech in 200+ voices, so you can narrate presentations or create audiobooks without recording anything yourself.
Mistral released Voxtral, which transcribes your 30-minute meetings with better accuracy than Whisper, while also answering questions directly from the audio—starting at $0.001 per minute (download or try the API).
Machined researches and writes fact-rich content for you by scraping search results or your chosen sources, complete with citations you can verify before publishing, and interlinks all the articles together for SEO (more on V3).
TestSprite automatically tests your software and validates AI-generated code, catching bugs before they reach production with zero manual testing required—free version available.
Velocity tests your Figma designs with AI users first—see where they click, what confuses them, then fix before human testing.
ZeroEntropy delivers personalized search results that learn from your queries and get smarter over time (raised $4.2M).
Here’s how to use Claude Code + Docker to code with AI on your computer in a containerized environment.
Mozart AI is Cursor for music; create music by talking to an AI co-producer that generates chord progressions, melodies, and loops in any artist's style (fun demo vid)

See our top 51 AI Tools for Business here!

Around the Horn.

This is Runway’s model that lets you act out a scene, then generate footage over it.

Apple's open-source machine learning framework, MLX, was updated to support a CUDA backend, allowing ML code written for Apple Silicon to run directly on Nvidia GPUs.
Ex-OpenAI CTO Mira Murati's startup Thinking Machines raised $2B at $12B valuation despite having no products, with A16z leading the massive seed round.
OpenEvidence raised $210M for its doctor-focused AI assistant platform, with 40% of US doctors adopting the free AI search tool by July 2025.
Google began testing AI-generated summaries in its Discover feed, which compile information from multiple sources and threatens publisher traffic.
Nextdoor launched a complete app redesign with AI recommendations and local news integration.
A new report found that 75% of companies in the S&P 500 now list AI as a risk in their risk disclosures, covering everything from ethical concerns, operational disruptions, regulatory scrutiny, and security.
A survey of 1,300+ US managers found that 60% now use AI for personnel decisions—and 78% of those use it to decide on raises.

For the latest AI deep dives, check out our Explainer articles here!