Welcome to P3 Media’s AI Commerce Brief, your daily update on the AI and commerce stories shaping how companies build, sell, and grow. It’s Wednesday, September 23. Let’s get into it.
Anthropic has introduced Claude Opus 5.5, the first model in its new Claude 5.5 family.
The company says the model performs at the level of Claude Fable 5.1 on most work, while costing 40 percent less to run than Opus 5.
Anthropic lists pricing at $4 per million input tokens and $20 per million output tokens. It also says Opus 5.5 generates output more than 30 percent faster than Opus 5.
The commercial point is not just another benchmark cycle.
Long-running coding, research, and business agents are expensive to operate. A model that preserves high-end capability while lowering serving cost can change which workflows are practical at scale.
Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks.
Watch for independent evaluations, and for how quickly enterprise platforms make the new family available.
OpenAI is pushing on the same cost curve.
It released GPT-6 Sol and GPT-6 Luna, with both models available in ChatGPT Work and Codex for eligible paid plans, and through the OpenAI API. They are not yet available in regular Chat.
For standard API usage, OpenAI prices Sol at $2 per million input tokens and $10 per million output tokens. Luna is priced at 10 cents for input and 50 cents for output.
OpenAI says that is 50 percent lower than the promotional pricing of their GPT-5.6 counterparts.
For enterprise buyers, the practical takeaway is to test model routing instead of assuming one model should handle every task. Expensive reasoning can be reserved for the steps that need it, while lower-cost models handle routine volume.
In China, Alibaba is pairing model ambition with its own compute strategy.
The Associated Press reports that Alibaba unveiled the Zhenwu V900 chip and described plans to train a new model at a scale of 5 to 10 trillion parameters.
Chief executive Eddie Wu called the V900 China’s most powerful AI chip and said it can deliver three times the performance of Alibaba’s prior Zhenwu M890.
Those are company claims, and outside validation will matter.
Alibaba also said its latest Qwen3.8-Max model has 2.4 trillion parameters, and that it expects to have more than 20 gigawatts of computing capacity by 2032.
The strategic signal is clear: Alibaba wants tighter control across chips, cloud infrastructure, and models.
Watch whether it can secure enough hardware and data center capacity to execute that plan.
In social commerce, Best Buy says it will launch on TikTok Shop in late October with nearly 10,000 products.
The assortment will include categories such as headphones, cameras, small appliances, laptops, and accessories. Customers will be able to buy inside TikTok, with Best Buy fulfilling the orders.
Eligible purchases can be returned in a store or by mail.
The retailer is also connecting creators and advertisers to the storefront. Creators can feature products in shoppable content, while Best Buy Ads partners can use creator content in campaigns tied to Best Buy’s first-party data and measurement.
For brands, that makes TikTok less of an awareness-only channel and more of a measurable path from discovery to purchase.
Amazon is also putting AI deeper into physical commerce.
The company says it will invest another $1.9 billion in its Delivery Service Partner program in 2027.
It is using more advanced AI techniques to identify driving trends linked to accidents, while adding surround-view cameras, dynamic hazard alerts, and routing that adjusts for real-world conditions.
Amazon says its Wellspring mapping technology has identified 202 million parking locations since 2024, along with 2.8 million building entrances and 85,000 mailrooms and lockers.
It also expects more than 20,000 Smart Delivery Glasses devices to be in the field by the end of 2027.
For commerce operators, the message is that last-mile advantage increasingly combines labor, hardware, software, and proprietary location data.
What matters next is evidence.
Watch independent testing of the new models, Alibaba’s ability to turn scale targets into deployed capacity, Best Buy’s holiday conversion and return patterns on TikTok Shop, and Amazon’s rollout of its delivery technology.
That’s your AI Commerce Brief for today. Thanks for listening.
Anthropic has introduced Claude Opus 5.5, the first model in its new Claude 5.5 family.
The company says the model performs at the level of Claude Fable 5.1 on most work, while costing 40 percent less to run than Opus 5.
Anthropic lists pricing at $4 per million input tokens and $20 per million output tokens. It also says Opus 5.5 generates output more than 30 percent faster than Opus 5.
The commercial point is not just another benchmark cycle.
Long-running coding, research, and business agents are expensive to operate. A model that preserves high-end capability while lowering serving cost can change which workflows are practical at scale.
Anthropic says Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks.
Watch for independent evaluations, and for how quickly enterprise platforms make the new family available.
OpenAI is pushing on the same cost curve.
It released GPT-6 Sol and GPT-6 Luna, with both models available in ChatGPT Work and Codex for eligible paid plans, and through the OpenAI API. They are not yet available in regular Chat.
For standard API usage, OpenAI prices Sol at $2 per million input tokens and $10 per million output tokens. Luna is priced at 10 cents for input and 50 cents for output.
OpenAI says that is 50 percent lower than the promotional pricing of their GPT-5.6 counterparts.
For enterprise buyers, the practical takeaway is to test model routing instead of assuming one model should handle every task. Expensive reasoning can be reserved for the steps that need it, while lower-cost models handle routine volume.
In China, Alibaba is pairing model ambition with its own compute strategy.
The Associated Press reports that Alibaba unveiled the Zhenwu V900 chip and described plans to train a new model at a scale of 5 to 10 trillion parameters.
Chief executive Eddie Wu called the V900 China’s most powerful AI chip and said it can deliver three times the performance of Alibaba’s prior Zhenwu M890.
Those are company claims, and outside validation will matter.
Alibaba also said its latest Qwen3.8-Max model has 2.4 trillion parameters, and that it expects to have more than 20 gigawatts of computing capacity by 2032.
The strategic signal is clear: Alibaba wants tighter control across chips, cloud infrastructure, and models.
Watch whether it can secure enough hardware and data center capacity to execute that plan.
In social commerce, Best Buy says it will launch on TikTok Shop in late October with nearly 10,000 products.
The assortment will include categories such as headphones, cameras, small appliances, laptops, and accessories. Customers will be able to buy inside TikTok, with Best Buy fulfilling the orders.
Eligible purchases can be returned in a store or by mail.
The retailer is also connecting creators and advertisers to the storefront. Creators can feature products in shoppable content, while Best Buy Ads partners can use creator content in campaigns tied to Best Buy’s first-party data and measurement.
For brands, that makes TikTok less of an awareness-only channel and more of a measurable path from discovery to purchase.
Amazon is also putting AI deeper into physical commerce.
The company says it will invest another $1.9 billion in its Delivery Service Partner program in 2027.
It is using more advanced AI techniques to identify driving trends linked to accidents, while adding surround-view cameras, dynamic hazard alerts, and routing that adjusts for real-world conditions.
Amazon says its Wellspring mapping technology has identified 202 million parking locations since 2024, along with 2.8 million building entrances and 85,000 mailrooms and lockers.
It also expects more than 20,000 Smart Delivery Glasses devices to be in the field by the end of 2027.
For commerce operators, the message is that last-mile advantage increasingly combines labor, hardware, software, and proprietary location data.
What matters next is evidence.
Watch independent testing of the new models, Alibaba’s ability to turn scale targets into deployed capacity, Best Buy’s holiday conversion and return patterns on TikTok Shop, and Amazon’s rollout of its delivery technology.
That’s your AI Commerce Brief for today. Thanks for listening.
