AI price war: Opus 5.5, GPT-6 Sol, and Qualcomm chips running pocket-sized agents

In less than 24 hours, the artificial intelligence market was shaken up once again. Anthropic and OpenAI launched cheaper models almost simultaneously, Qualcomm unveiled chips designed to run agents on phones, and OpenAI reinforced its focus on content creators with new hires from Patreon. Three distinct signals, one message: the race is no longer measured solely by benchmarks, but by cost, hardware, and new business opportunities.

Anthropic and OpenAI ignite a price war

On September 22, 2026, Anthropic released Claude Opus 5.5. Approximately 90 minutes later, OpenAI responded with GPT-6 Sol and GPT-6 Luna. This was no coincidence: both companies drove prices down while continuing to sell more capacity.

According to TechCrunch and Simon Willison's analysis, Opus 5.5 lowers the buy-in cost to $20 per million tokens (down from $25) and the buy-in cost to $4 per million. Anthropic claims it approaches the performance of its flagship Fable model in many tasks, with increased speed and improved caching efficiency. It also highlights better results in programming and knowledge work, and says it's their best model yet in internal alignment tests.

OpenAI, for its part, halved the price of Sol and Luna compared to the 5.6 series. GPT-6 Luna settled at around $0.10 per million tokens for entry and $0.50 for exit, an aggressive level for high-volume tasks. Sol targets complex work such as code; Luna, for summarizing, extracting, and fast queries. OpenAI claims that Sol makes approximately half the factual errors of its predecessor and achieves near-Astra reliability at a lower cost.

The paradox is clear. Just days earlier, the CEOs of Anthropic and OpenAI themselves had spoken of slowing the pace of AI development. At the same time, both labs are competing to steal clients from each other and defend their profit margins against cheaper Chinese models. AFP aptly summarizes the climate: more confidence in the rhetoric, more commercial pressure on the product. Opus 5.5 applies similar safeguards to Fable in cybersecurity, biology, and model design, but Anthropic admits that the system sometimes suspects it is being evaluated, which complicates predicting its actual behavior.

For developers and businesses, the immediate effect is practical. Building agents, support bots, or code pipelines no longer always requires the most expensive model. The bottleneck is shifting from raw price to orchestration, data quality, and risk management.

Qualcomm puts agentic AI inside the smartphone

While labs battle it out in the cloud, Qualcomm has focused on the device. At its Snapdragon Summit, it unveiled two top-of-the-line processors: the Snapdragon 8 Elite Gen 6 and the Snapdragon 8 Elite Extreme Gen 6, with a clear emphasis on artificial intelligence capabilities.

TechCrunch details that both chips incorporate sensor hubs capable of running small models of up to 200 million parameters locally. This allows for personalized scribing, voice recognition, and the building of usage memory to suggest automations. Qualcomm claims the chip can support a complete voice agent, both incoming and outgoing, without always relying on the cloud.

The Extreme variant goes even further: it can run a mixture-of-experts model of 30 billion parameters on the phone, activating only a portion depending on the task. For reference, Apple presented a 20 billion MoE at WWDC as its most advanced third-generation foundation model. In terms of camera and video, Extreme promises 8K at 60 fps, 4K at 240 fps for slow motion, and the APV codec for professional-grade recording. There are also voice enhancements: vocal boosting, noise reduction, and "voice bubble" technology to isolate the user during calls.

Motorola has already announced the Signature 27 with the Extreme Gen 6 processor, expected to be available this year. Qualcomm's message is clear: many people will use AI from their mobile devices, not from dedicated gadgets. If these pocket-sized AI agents work well, it reduces reliance on expensive APIs for everyday tasks and increases the pressure on privacy, battery life, and on-device model updates.

OpenAI hires the team that invented creator subscriptions

The third piece of news today isn't a model, but a product move. The Verge reported that OpenAI has hired three former Patreon executives to lead its creator strategy. Sam Yam, co-founder and former chief technology officer of Patreon, will head Creator Product. He's joined by Drew Rowny (product) and Shannon Ma (engineering), both of whom left Patreon in recent weeks.

Yam wrote on X that they will build alongside creators and share early access to new tools. He asked people to pay attention to OpenAI DevDay, scheduled for Tuesday, September 29, and linked to a message from Sam Altman hinting at another announcement this week.

The market reading is straightforward. Patreon defined the template for creator subscriptions. If OpenAI integrates that experience into ChatGPT, Codex, or an agent marketplace, it can open a revenue stream distinct from enterprise subscriptions and the API. For Patreon, Yam's departure comes after a 20% staff reduction in January and a feature upgrade plan. For OpenAI, the timing aligns with DevDay and the narrative that AI not only answers questions but can also support communities, payments, and creative tools.

What connects these three news stories?

The three stories from September 22nd and 23rd, 2026, paint the same picture. In the cloud, Anthropic and OpenAI compete on performance per dollar. In the pocket, Qualcomm tries to put the agent on the chip. In the business world, OpenAI seeks to monetize creators with the team that already built that market.

None of these developments guarantee a single winner. They do, however, point to a stage where cost, local implementation, and new products will be just as important as the largest current model. Anyone building AI products should consider all three layers simultaneously: cheaper APIs, locally capable hardware, and distribution channels to real-world communities.

Sources

• TechCrunch: Anthropic releases Opus 5.5; OpenAI launches GPT-6 Sol and Luna; Qualcomm Snapdragon AI chips
• Simon Willison: Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war (Sep 22, 2026)
• AFP / The Star: Anthropic, OpenAI release cheaper AI even as safety fears grow
• The Verge: OpenAI nabs key Patreon execs ahead of upcoming announcement