Retail AI Report: New Models & Cheaper Prices

Anthropic's Claude Opus 5 Makes Frontier AI a Line-Item Retailers Can Afford

Artificial Intelligence

Anthropic released Claude Opus 5 on July 24, claiming it delivers nearly all the intelligence of its flagship Fable 5 model at roughly half the cost, holding pricing at $5 per million input tokens and $25 per million output tokens. The company emphasized efficiency gains — fewer tool calls, faster completion of enterprise tasks — and made it the default model across its Max and Pro tiers. It is positioned as an everyday workhorse for routine enterprise tasks. It is estimated to be 25% more efficient in token usage than Opus 4.8.

Why it matters: The total cost of ownership of an AI system is what determines whether it graduates from pilot to production, and retail runs at enormous volume — millions of product descriptions, customer queries, and forecasting runs. A frontier-class model that is cheaper and completes tasks in fewer steps directly lowers the per-transaction cost of AI in merchandising, service, and supply-chain agents. Leaders evaluating AI roadmaps should re-run the ROI math on use cases that were previously too expensive to scale.

Source: VentureBeat

Moonshot AI's $35 Billion Repricing Signals a Frontier Racing Toward Commodity

Artificial Intelligence

China's Moonshot AI closed a $3.5 billion raise — well above its original $1–2 billion target — vaulting to a $35 billion valuation on the strength of its Kimi K3 model. Backers included China's National Artificial Intelligence Industry Investment Fund, also a DeepSeek investor. The episode shows how a single strong release can reprice an entire lab in weeks.

Why it matters: Capable, lower-cost Chinese frontier models exert steady downward pressure on the pricing of the US labs whose APIs sit inside most retail AI stacks — good news for retailers' input costs over time. The deeper lesson is commoditization: no model vendor holds a durable lead for long. Retailers should build model-agnostic architectures that let them swap providers as price and performance shift, rather than hard-wiring their systems to a single API.

Source: Bloomberg

Google's Gemini 3.6 Flash Brings "Computer Use" to the Cheap Tier

Artificial Intelligence

Google launched Gemini 3.6 Flash on July 22 at $1.50/$7.50 per million tokens, roughly 17% more token-efficient than its predecessor and — notably — with "Computer Use" built in, meaning the model can directly operate software interfaces. Google also shipped an even cheaper Flash-Lite tier, continuing an extraordinary pace of model launches through July. "Computer Use" is the capability that lets an AI click, type, and navigate applications the way a human employee would.

Why it matters: Embedding agentic control into a low-cost model pushes real automation of back-office work within reach — think catalog updates, order reconciliation, and store-ops paperwork handled by an AI operating existing systems, no custom integration required. Paired with flash-tier pricing, high-volume tasks like product tagging and customer-service triage become economically viable at full scale. Retailers should identify repetitive interface-driven workflows now, as they are the first candidates for agentic automation.

Source: AIToolsRecap

AI Steps Onto the Sales Floor as Store Operations Become the New Frontier

Retail

A VoCoVo survey of 250 US retail executives and 500 consumers found AI moving decisively beyond e-commerce and marketing into physical store operations — associate productivity, shelf availability, and in-aisle customer assistance. Until recently, retail AI was concentrated in e-commerce and marketing; the survey points to growing adoption in physical stores, where most sales still occur.

Why it matters: The store is the largest, least-digitized part of most retail P&Ls, so AI that improves associate efficiency and on-shelf availability targets the biggest pool of unrealized value. But physical-store AI succeeds or fails on frontline adoption, not model quality — meaning change management and staff training matter as much as the technology. Executives should revisit store-level tech budgets and build adoption plans before greenlighting deployments.

Source: WWD / Sourcing Journal

Meta Embeds Agentic AI Into Facebook, Reshaping Social Commerce

Retail

Meta made its Muse Spark 1.1 agentic model available to consumers and rolled out new Facebook features built on it. The model carries a 1-million-token context window, can operate across desktop, browser, and mobile, and delegates work to parallel subagents. The consumer rollout follows Meta's first paid developer API for the model.

Why it matters: Putting agentic AI directly into Facebook and Instagram surfaces means shoppers can ask product questions, get recommendations, and complete purchases inside conversations — collapsing discovery and checkout into a single chat. Brands that depend on Meta for reach need clean, structured product data and agent-ready storefronts, because conversational selling rewards machine-readable catalogs over polished web pages. This is a preview of how discovery may work when the interface is an agent, not a feed.

Source: SiliconANGLE

Kling AI's $2.8 Billion Raise Signals Generative Video's Move Into the Marketing Stack

Artificial Intelligence

Kling AI raised roughly $2.8 billion at an $18 billion post-money valuation, backed by Alibaba and Tencent, in one of the largest single rounds ever for an AI video generation platform. The capital positions Kling for greater independence from parent Kuaishou and reflects surging demand for AI video among creators and brands.

Why it matters: Product video, ads, and social creative have long been among the most expensive line items for retail marketing teams; well-funded, fast-improving video generation collapses that cost. CMOs should expect generative video to become a standard tool for producing personalized creative across channels and thousands of SKUs — turning what was once a per-shoot expense into a scalable software cost. The competitive edge will shift toward teams that master prompt-and-iterate creative workflows.

Source: Bloomberg via Quasa

What to Watch: Watch for the price competition among frontier and flash-tier models to keep compressing API costs through the back half of 2026 — a tailwind for any retailer scaling AI. Keep an eye on the Washington debate over open-source AI restrictions, which could narrow retailers' ability to fine-tune models on proprietary data in-house. And track how quickly "Computer Use" and agentic checkout move from demos to deployed workflows, because that is where the next wave of measurable retail returns will surface.