OpenAI CEO Sam Altman: OpenAI Aims to Provide the Best Models at Every Price Point for APIs
Sam Altman stated that OpenAI aims to provide the best models at every price point through its API, excelling in every modality including text, code, images, and video; developers will then come up with good ideas to build upon, with the best ideas coming from outside.
This statement breaks down the platform strategy into two layers: the company is responsible for the combination of intelligence and pricing on the Pareto frontier, as well as full modality coverage; application layer creativity is outsourced to the callers. A tiered pricing structure has emerged: GPT-6 Astra for complex end-to-end tasks, approximately $10 per million tokens for input and $50 for output; Sol for everyday judgments and programming, around $2 and $10; Luna for high-throughput, cost-sensitive tasks, about $0.10 and $0.50, with caching discounts.
The previous 5.6 series had already seen price reductions: Luna's input and output were once reduced by about 80%, Terra by about 20%, and Sol added a fast mode for a higher unit price to achieve speeds up to about 2.5 times faster. The official source of efficiency is attributed to simultaneous improvements in models, inference systems, and agent connection layers, along with distillation to compress frontier capabilities into cheaper smaller models. Real-time voice, image generation, and Sora video are priced separately by the minute or second, forming a shelf of modalities beyond text.
"The best at every price point" directly addresses the cost-performance narrative of open-source and low-cost closed-source competitors. He has previously stated the goal of providing optimal intelligence-price options at every point on the Pareto frontier, asserting that it must be both the strongest and the cheapest. Outsourcing creativity positions the API as a utility, leaving differentiation to the product judgments of the callers.
In market mechanisms, this is a cloud inference price war combined with platform revenue sharing: buyers are developers who need to switch models for different tasks, while sellers use tiered pricing to lock in the entire usage curve from experimentation to production. Funds flow from application budgets into token bills, then into computing power and distillation pipelines. Beneficiaries are teams that can route tasks to Astra, Sol, and Luna; those under pressure are competitors with either only expensive models or only cheap but unstable models. The event-driven aspect is set by the top management, not the day the new price list takes effect.
The denser the pricing table, the higher the migration costs: once developers have written routing according to three tiers, they will not rewrite the entire chain just to save a little money.
Source: Public Information
ABAB AI Insight
Altman has rewritten OpenAI from "a flagship model" to a shelf: winning at every price point, winning in every modality, with ideas left for others to conceive. This is a classic revenue-sharing model for platform companies in the agent era—retaining reasoning and multimodal infrastructure while offloading the uncertainty of application creativity to developers. The price differences between Astra, Sol, and Luna are not marketing tiers but productizing distillation capabilities: frontier models teach smaller models, which are then sold in volume.
The capital path is to use price reductions to secure lock-in. The Luna-level price drop brings experimental costs to negligible levels, Sol handles daily production, and Astra tackles high-difficulty bills. Money flows from the caller's monthly budget into continuous reasoning, rather than a one-time authorization. The motivation is to prevent developers from establishing default routes on low-cost open-source options. Resource mobilization relies on self-distillation of models, forming a cost curve internal cycle.
Analogies can be drawn to Amazon Cloud's instance tiers, NVIDIA's product tree from flagship cards to inference cards, and the big.LITTLE architecture of mobile chips. The industry phase has shifted from competing on who has the largest parameters to competing on who can provide acceptable intelligence density at every tier. The position is at the API layer: whoever becomes the default router collects tolls for all modalities.
Structural judgment pertains to the transfer of pricing power. The mechanism is: when model intelligence can be distilled, competition shifts from unique top models to the envelope of the entire price curve; whoever fills the curve first can make developers' code depend on their tier names. The best ideas come from outside, indicating that OpenAI has abandoned the fantasy of sharing in the application layer and instead collects utility fees.
ABAB News · Cognitive Laws
The platform aims to win the entire price curve, not just a flagship
Distillation turns top capabilities into utility that can be sold by volume
The best ideas are outsourced, allowing tolls to be collected longer