🛒 AWS launches AI Agent Marketplace next week 🛒
Amazon's move to centralize AI agent distribution

Amazon Web Services is set to launch an AI agent marketplace on July 15 at the AWS Summit in New York City, with Anthropic as a key partner, according to TechCrunch sources. The marketplace will allow startups to directly offer AI agents to AWS customers and enable enterprises to browse, install, and find AI agents from a single location, addressing the current distribution challenge where companies offer agents in silos. Anthropic, which already has Amazon's $8 billion backing and is reportedly in line for another multibillion-dollar investment, views AI's future primarily in terms of agents and already builds them in-house while enabling developers to create them via API.

🎬 Gemini adds Photo-to-Video Feature Powered by Veo 3 🎬
8-second clips with sound available to AI Pro and Ultra subscribers

Google has launched a photo-to-video capability in Gemini, allowing users to transform static images into 8-second video clips with AI-generated sound using the Veo 3 model. Available to Google AI Pro and Ultra subscribers in select regions, the feature began rolling out Thursday on web and will reach mobile throughout the week. Users can upload a photo, describe the desired scene and audio, and watch as Veo 3 creates MP4 videos at 720p in 16:9 format with sound effects, ambient noise, and speech.

Google reports over 40 million Veo 3 videos generated across Gemini and Flow in just seven weeks, with users creating everything from animated fairy tales to ASMR content. The feature includes visible and invisible SynthID watermarks to identify AI-generated content and employs extensive safety measures including restrictions on public figures and harmful content.

💻 Mistral AI Launches Devstral Medium and Upgrades Small 1.1 💻
New models for agentic development high performance

Mistral AI has introduced Devstral Medium and upgraded Devstral Small 1.1 in collaboration with All Hands AI, focusing on enhanced generalization for different prompts and agentic scaffolds. Devstral Small 1.1 maintains 24B parameters under Apache 2.0 license and achieves 53.6% on SWE-Bench Verified, setting a new state-of-the-art for open models without test-time scaling.

The model supports both Mistral function calling and XML formats for versatility across coding environments. Devstral Medium delivers 61.6% on SWE-Bench Verified and is available through API at competitive pricing, surpassing Gemini 2.5 Pro and GPT-4.1 for a quarter of the price.
Both models are available through API with devstral-small-2507 priced at $0.1/M input tokens and devstral-medium-2507 at $0.4/M input tokens.

INTERESTING TO KNOW

🏥 Google unveils MedGemma most capable open models for Health AI 🏥 

Google Research has announced new models in the MedGemma collection, calling them their "most capable open models for health AI development." The release includes MedGemma 27B Multimodal for complex electronic health record interpretation and MedSigLIP, a lightweight 400M parameter image encoder for medical classification and retrieval tasks.

MedGemma 4B scores 64.4% on MedQA medical knowledge benchmarks, with 81% of its chest X-ray reports judged by board-certified radiologists as accurate enough for similar patient management. The 27B variant achieves 87.7% on MedQA, performing within 3 points of DeepSeek R1 but at one-tenth the inference cost.

💼 Salesforce Hits 1 Million AI Agent Conversations 💼 

Agentforce platform reaches milestone since October launch, says CFO Salesforce has surpassed 1 million AI agent-customer conversations through its Agentforce platform, with agents autonomously resolving 85% of all customer queries, according to President and COFO Robin Washington.

The milestone comes less than a year after the October 2024 launch of Agentforce, an autonomous AI agent platform providing 24/7 AI-powered customer support.