Meta climbs the AI image leaderboard
Meta released Muse Image, an in-house AI image model ranking No. 2 on leaderboards behind only OpenAI's GPT Image 2, with a teased video model also performing strongly. Meanwhile, Beijing is considering restrictions on Chinese AI model exports, creating potential geopolitical reciprocity risks similar to U.S. export controls, while companies like DoorDash demonstrate that pairing multiple AI models significantly improves code review accuracy and cost-effectiveness.
Summary
Meta's Superintelligence Labs, led by Alexandr Wang, has launched Muse Image as the company's first in-house AI image generation model. The model debuted at No. 2 on Arena's text-to-image and editing leaderboards, trailing only OpenAI's GPT Image 2. Muse Image features agentic capabilities including web search and tool use, and can edit its own outputs for quality improvement. The model is available free within Meta AI and rolling out across Instagram and WhatsApp, with expansion to Facebook, Messenger, and Meta's ads platform planned later. Meta also previewed a Muse Video model that ranked No. 3 on Arena's leaderboard. This represents a significant strategic shift for Meta, which previously relied on partnerships with Midjourney and Black Forest Labs for creative AI capabilities.
On the geopolitical front, Beijing is reportedly discussing restrictions that would limit Chinese AI models' use outside of China. Commerce officials met with ByteDance, Alibaba, and Zhipu AI to discuss potential foreign-use limitations for their strongest models including Qwen, Doubao, and GLM-5.2. These discussions encompassed both closed and open-source models, startup funding limits, and penalties for proprietary model leaks. This move mirrors June's U.S. export controls on Anthropic's models, creating a reciprocal risk where Western users could lose access to increasingly popular and cost-effective Chinese models as quickly as Chinese regulators can implement restrictions.
In AI research, DoorDash published DashBench, an internal benchmark evaluating AI code reviewers against 105 past code changes. Their testing revealed that single models caught 20-30% of problems, while pairing two Claude models identified just over half. Most impressively, combining the open-source Kimi K2.6 with Claude Fable 5 caught approximately two-thirds of problems and 8 of 10 critical bugs at just $3.81 per review, demonstrating both cost efficiency and superior quality compared to single-model approaches. Additional developments include Anthropic extending Fable 5's availability in subscription plans until July 12, Microsoft reportedly running Excel and Outlook prompts on in-house MAI models to reduce Anthropic spending, and Figma acquiring the team behind AI agent platform Bud.
About this episode
PLUS: Run better 1-on-1s with employees using AI
Key Insights
- Meta shifted from outsourcing creative AI to Midjourney and Black Forest Labs to developing Muse Image in-house, achieving No. 2 ranking on leaderboards and gaining strategic control over AI capabilities integrated across its social platforms and advertising systems.
- Beijing's proposed restrictions on Chinese AI model exports would create reciprocal geopolitical risk, potentially cutting off Western users from increasingly popular low-cost models just as quickly as U.S. export controls affected Western companies, establishing a two-sided vulnerability in global AI access.
- DoorDash's benchmark found that pairing open-source Kimi K2.6 with Claude Fable 5 for code review caught two-thirds of problems at $3.81 per review, demonstrating that multi-model combinations can achieve superior quality and cost efficiency compared to premium single-model solutions.
- Open-source AI models have reached performance parity with premium alternatives in specific tasks, with Kimi K2.6 becoming serious contenders in production pipelines where cost savings and efficiency opportunities grow with each new release.
- The competitive landscape shows companies like Microsoft actively working to reduce reliance on third-party AI providers by developing in-house models and running inference on proprietary infrastructure, signaling a broader industry shift toward vertical AI integration.
Topics
Transcript
Good morning, {{ first_name | AI enthusiasts }}. Meta's Alexandr Wang was talking a big game days ago, saying the in-training 'Watermelon' model had pulled level with GPT-5.5 and promising an X user, "you'll like what we have cooking." Today, Meta’s kitchen served something even more surprising: Muse Image, which opens at No. 2 on the leaderboards with an equally strong video model teaser behind it — the kind of receipts that make the Watermelon hype even harder to dismiss. Reminder : Our next live workshop is today at 12 PM EST. Join and learn how to create a viral AI avatar content pipeline that covers topics, research, avatar and voice generation, and monetization strategies. RSVP here . Meta’s new…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Rundown AI
The part of Claude's brain nobody built
Anthropic researchers discovered 'J-space,' an internal workspace in Claude that wasn't explicitly programmed but emerged during training, functioning similarly to human conscious thought. The newsletter also covers Tencent's efficient Hy3 model, Kyutai's Rocket League world model, and practical AI applications including mobile app development with Replit.
Meta sizes up GPT-5.5 with 'Watermelon'
Meta's upcoming 'Watermelon' AI model reportedly matches OpenAI's GPT-5.5 performance while still in training, though the AI frontier continues advancing with competing models like Mythos and Fable. The newsletter covers Meta's $145B AI investment progress, practical AI applications from users, and emerging AI tools across education, coding, and consumer devices.
Altman invites Washington inside the AI industry
Sam Altman proposes a U.S.-led international AI safety forum modeled after the IAEA, with reports suggesting OpenAI discussed offering the government a 5% equity stake. Meanwhile, Meta's smart glasses capture 80% of the market but still fall short of replacing phones, and specialized AI models trained on expert data outperform frontier models at significantly lower costs.
Anthropic's Fable returns worldwide
Anthropic's Fable 5 AI model has returned to public access after a 3-week government-mandated shutdown, now featuring enhanced safety filters and requiring U.S. government pre-release access to future models. The newsletter also covers Meta's cloud compute rental plans, AI design tools, and new benchmarks showing Fable 5 achieving record performance on freelance task automation at 16.1% professional quality.
Sonnet 5 ships as Washington frees Fable
Anthropic launched Sonnet 5, a mid-tier model with improvements in agentic coding and reasoning, though its release is overshadowed by the imminent return of the more powerful Fable 5 and Mythos 5 models after export control restrictions were lifted. The newsletter also covers new AI tools from Google, Anthropic's Claude Science platform, and emerging developments in AI hardware and infrastructure.