• AI Valley
  • Posts
  • An open-source model outperforms GPT-5

An open-source model outperforms GPT-5

PLUS: Perplexity to pay Snapchat $400M to power search

Author

Barsee
07 Nov

Together with

Howdy, it’s Barsee.

Happy Friday, AI family, and welcome to another AI Valley edition. This issue takes 4 minutes to read.

Today’s climb through the Valley reveals:

  • Kimi-K2 Thinking outperforms GPT-5

  • Perplexity to pay Snapchat $400M to power search

  • Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

VOICES.COM

Image Credit: Voices.com

The best AI voices don’t feel robotic, because they’re not. They’re powered by real, talented humans. Customers want a voice AI experience that feels real and authentic, and for voice AI to truly represent your company or your product, it needs to sound natural, distinct, and unmistakably on-brand.

Join Voices on November 13th at 1pm ET to hear how Voices is helping brands discover and source professional voices to bring their AI experiences to life.

You’ll also hear from BMW on how they create authentic in-car AI experiences with voice and how Voices helped them do it.

*This is sponsored

THROUGH THE VALLEY

Moonshot AI has released Kimi K2 Thinking, a trillion-parameter open-source model that just beat OpenAI’s GPT-5 and Anthropic’s Claude Sonnet 4.5 in key reasoning and coding benchmarks.

The model’s architecture activates 32 billion parameters per inference, supports 256k-token context windows, and can autonomously perform 200–300 sequential tool calls without human input.

Despite its size, it’s fast and cheap, costing just $0.15 per 1M tokens and $2.50 per 1M output tokens, nearly 10× cheaper than GPT-5 and 20x cheaper than Sonnet 4.5.

Moonshot built K2 to think like a human. Each answer shows every step of logic before reaching a conclusion, making its reasoning transparent and easy to audit.

Here are some of its benchmark results:

Source: Moonshot AI

  • Humanity’s Last Exam: 44.9 % (with tools enabled)

  • BrowseComp: 60.2 % (web reasoning + search)

  • SWE-Bench Verified: 71.3 % (coding + tool use)

Across these tests, K2 Thinking consistently outperforms GPT-5, Claude Sonnet 4.5, and xAI’s Grok-4, setting a new bar for open-source reasoning models.

The release includes APIs for chat, reasoning, and automation. It’s fully open-source under a Modified MIT License. The only rule: if your app serves more than 100 million users or earns over 20 million dollars a month, you must credit “Kimi K2” in the interface.

Why does it matter?

For the first time, an open-weight AI model has not only caught up with but surpassed the world’s leading proprietary systems, and it’s free. While OpenAI pours trillions into chips and datacenters, Moonshot is proving that clever architecture > infinite compute. If enterprises can now get GPT-5-level reasoning at 1/10 the cost, the case for paying premium prices for closed systems starts to crumble.

AI search startup Perplexity will integrate its “answer engine” directly into Snapchat over the next year in a paid deal. Under the agreement, Perplexity will pay Snap $400 million over one year, through a combination of cash and equity

For Snap, this could be more than an AI experiment; it’s a strategic pivot. The company’s built-in chatbot, My AI, has gained users but struggled to stand out against Meta’s expanding AI features. Partnering with Perplexity could give Snap a fresh tech edge, helping reposition it from a social app into an AI-powered discovery hub.

For Perplexity, the partnership is massive: Snap’s 943 million-strong user base offers a direct line to a notoriously hard audience to capture (young users who increasingly get information from messaging apps instead of search engines).

Why does it matter?

The deal signals a generational shift in how people search the web. Instead of typing queries into Google, millions of younger users could soon be asking questions inside Snapchat, powered by Perplexity’s AI. It’s a bold, costly bet (nearly a third of Perplexity’s $1.5 billion funding), but if it works, it could cement Perplexity as the default search engine for many Snapchat users.

TRENDING TOOLS

  • Kimi K2 Thinking > Moonshot AI’s new open-source agent that pushes state-of-the-art reasoning to new heights

  • Firecrawl > Instantly extract a brand’s entire DNA from any website (colors, logos, frameworks, and more)

  • Cotera > Build AI agents directly in chat, connect them to hundreds of apps, and automate repetitive work

  • Stream Ring > A sleek AI smart ring that lets you record voice notes with just a whisper

  • TryCaddy > Control your computer and every app using only your voice

  • Manus 1.5 > A general-purpose AI agent that builds full-stack web apps through natural conversation

  • Orgo > Virtual computers for AI agents, they can now create files, browse the web, and run desktop apps autonomously

  • Gemini Deep Research > Adds new connectors for Gmail, Drive, Docs, and Chat to supercharge research workflows

  • MatterAl > A self-improving code intelligence layer that reviews, optimizes, and verifies code against your company’s architecture, security, and performance standards.

THINK PIECES / BRAIN BOOST

THE VALLEY GEMS

What’s trending on social today:

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”.