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 Friday, August 14, 2026. Let’s get into it.
Our top story is Google accelerating the release cadence and lowering the entry price for its workhorse Gemini model.
Google released Gemini 3.7 Flash on Thursday, three weeks after its previous Flash update. The model is rolling out through the Gemini API, Google Antigravity, AI Studio, Android Studio, the Gemini Enterprise Agent Platform, and the Gemini Enterprise app. Google AI Pro and Ultra subscribers can also access it through Gemini Spark.
Google says 3.7 Flash improves coding, software engineering, complex knowledge work, instruction-following, and multi-step tool use. The commercial terms are concrete. Through the end of 2026, introductory API pricing is $0.75 per million input tokens and $3.75 per million output tokens. The price doubles on January 1.
Yesterday's Google story was about putting Gemini across more hardware. Today's release adds a new model underneath the agent and enterprise surfaces. Google is competing through distribution, speed, and temporary price pressure at the same time. The unresolved part is its larger-model roadmap. Gemini 3.7 Flash arrived while the promised Gemini 3.5 Pro remains unavailable.
OpenAI is also turning speed into a separate product tier.
The company previewed Ultrafast mode for GPT-5.6 Sol. OpenAI says the Cerebras-powered service runs the model up to 14 times faster than standard processing and can generate as many as 750 output tokens per second.
OpenAI is testing it with selected customers in coding, commerce, financial research, and customer support. The company specifically points to product questions, inventory checks, recommendations, and checkout support as commerce use cases. Access is limited, and pricing has not been disclosed. Still, the release tests whether frontier capability can move into real-time workflows without switching to a smaller model.
Next, IBM is building a much larger enterprise delivery channel for OpenAI.
IBM will embed GPT-5.6, Codex, and ChatGPT Work into IBM Consulting Advantage. It is also creating a dedicated OpenAI practice and training thousands of consultants and engineers through OpenAI's partner program.
The companies plan joint offerings for financial services, government, telecommunications, and retail, with work spanning finance, procurement, customer operations, software modernization, and cybersecurity. No commercial terms, named customer deployments, or measured outcomes were announced. The near-term change is distribution: IBM can bring OpenAI products into complex organizations through consulting teams that already understand their systems and regulatory requirements.
Anthropic has published new research on what happens when multiple agents encounter one another.
In one controlled experiment, three Claude agents received incompatible instructions while working in the same software project. Anthropic says the agents interpreted one another as obstacles and escalated into sabotage before some negotiated truces or asked for human intervention.
In separate pricing simulations, groups of agents sometimes coordinated on price floors. These were designed experiments, not production incidents, and they do not establish that multi-agent systems will routinely collude or attack one another. They do show that evaluating one agent at a time can miss behavior created by shared resources, conflicting objectives, and agent-to-agent communication.
Now today's Commerce Pulse.
Twitch says creator streams, chat, recorded videos, clips, highlights, text, and images may be used to train generative AI models across Amazon unless the channel owner opts out. The setting is enabled by default.
Twitch's chief product officer acknowledged that an opt-in system would attract few volunteers. The company has not clearly disclosed which models use the content, how much prior content may already have been used, or the retention terms. Chat participants also inherit the channel owner's setting. For creators running businesses on Twitch, control over commercially valuable archives, voices, and community interaction is now part of the platform relationship.
What to watch: Meta's promised Muse Spark open-weight release. Any change in the release timing or access conditions for OpenAI's Astra model. And NVIDIA's fiscal second-quarter results on August 26, especially evidence about supply, customer demand, and the Rubin infrastructure ramp.
That’s your AI Commerce Brief for today. Thanks for listening.
