AI News Daily 2026-09-24
Contents
- OpenAI cut API prices by 50% versus its prior generation while launching two new lightweight models, GPT-6 Sol and GPT-6 Luna, across ChatGPT Work, Codex, and the API.
- Anthropic introduced a limited-preview “Claude Marketplace” that lets enterprises discover partner-built Claude tools and pay for them using their existing Anthropic spend commitments.
- Qualcomm unveiled new Snapdragon smartphone chips built on TSMC’s 2nm process for on-device AI, alongside a datacenter chip architecture, Dragonfly, that it claims delivers up to eight times the tokens-per-watt efficiency of GPUs.
- The scale of the OpenAI price cut means enterprises relying on frontier-class models should revisit their AI cost assumptions now rather than at contract renewal.
- Anthropic’s marketplace move signals that leading AI vendors are positioning themselves as procurement hubs for third-party AI tools, not just as model providers.
01 OpenAI unveils GPT-6 Sol and GPT-6 Luna with a 50% API price cut
Published: 2026-09-22
Facts
OpenAI announced two new lightweight models, GPT-6 Sol and GPT-6 Luna, trained using an approach similar to the one used for GPT-6 Astra. Both models are now available through ChatGPT Work, Codex, and the OpenAI API. API pricing was cut 50% versus the recommended pricing for GPT-5.6: GPT-6 Sol is priced at $2 per million input tokens and $10 per million output tokens, while GPT-6 Luna is priced at $0.10 per million input tokens and $0.50 per million output tokens.
Background
Sol and Luna extend the training approach OpenAI used for its GPT-6 Astra model into smaller, lower-cost tiers, aimed at high-volume production workloads rather than frontier-only use cases.
Implications
A 50% price cut on capable models lowers the cost of embedding AI into everyday business workflows and is likely to push many companies to re-run their cost projections for API-based AI usage. It also raises competitive pressure on rival model providers to compete on price as well as capability.
02 Anthropic launches Claude Marketplace for enterprise partners
Published: 2026-09-23
Facts
Anthropic announced the Claude Marketplace, a limited-preview hub where businesses can discover plugins, agents, and services built by partners on top of Claude. Initial partners include GitLab, Harvey, Lovable, Replit, Rogo, and Snowflake. Enterprises can apply their existing Anthropic spend commitments toward purchasing tools from these partners.
Background
The marketplace extends Anthropic’s enterprise offering beyond the Claude models themselves, bundling a curated set of partner products already built on the platform — from coding tools (Replit) to legal AI (Harvey) to data platforms (Snowflake).
Implications
By consolidating procurement into a single storefront tied to existing spend commitments, Anthropic could reduce the time and cost enterprises spend evaluating and contracting with third-party AI vendors, while also deepening customer commitment to the Claude ecosystem.
03 Qualcomm debuts AI-focused smartphone chips and the Dragonfly datacenter architecture
Published: 2026-09-22
Facts
Qualcomm announced the Snapdragon 8 Elite Gen 6 and Snapdragon 8 Elite Extreme Gen 6, manufactured on TSMC’s 2nm process, which the company says can process up to one million tokens per day directly on the device. The chips are expected to ship in premium smartphones from Motorola, Xiaomi, and ZTE. Qualcomm also unveiled Dragonfly, a new datacenter chip architecture the company says delivers up to eight times the tokens-per-watt efficiency of GPUs.
Background
The announcement arrives amid an industry-wide memory shortage and rising demand for both on-device and datacenter AI compute, positioning Qualcomm to compete on two fronts at once: mobile silicon for on-device inference and dedicated chips for large-scale AI datacenters.
Implications
Simultaneous progress on-device and in the datacenter could reshape the cost structure of AI deployment at both ends — cheaper local inference on flagship phones, and, if Dragonfly’s efficiency claims hold up in practice, a new efficiency benchmark for AI datacenter chips that challenges incumbent GPU makers.
CNBC: Qualcomm releases Android chip built for AI amid memory shortage · TechCrunch: Qualcomm launches two new smartphone chips with emphasis on AI
Editor’s Note
Today’s three stories trace a common thread: every layer of the AI stack is getting cheaper and more consolidated at the same time. OpenAI’s 50% price cut on newly launched models (story 1) shows frontier labs competing as much on cost as on raw capability. Anthropic’s move to bundle partner tools into a single marketplace (story 2) suggests leading providers are trying to become the procurement layer for enterprise AI, not just the model layer. And Qualcomm’s parallel push into on-device and datacenter chips (story 3) shows the hardware side racing to keep up with both ends of that demand curve. Taken together, the direction for buyers is consistent across all three items: AI capability is becoming cheaper and more bundled, and procurement decisions made today may look expensive within a matter of months.