OpenAI Splits GPT-6 API Into Three Pricing Tiers and Lists ChatGPT Enterprise on AWS Marketplace
OpenAI's November 2026 API price sheet formalises GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna with 90% prompt caching discounts while opening ChatGPT Enterprise procurement through AWS.

By Clara Tan, Artificial Intelligence
, updated 7 min read
OpenAI has formalised a three-tier pricing and model architecture across its GPT-6 family, pairing steep cost reductions for cached and mid-tier agent workloads with a new enterprise procurement channel inside the Amazon Web Services Marketplace.
Published in OpenAI's updated November 2026 developer documentation alongside its 9 November AWS Marketplace launch for ChatGPT Enterprise, the structure reveals how the company is defending developer margins against lower-cost rivals while simplifying procurement for large corporate IT budgets.
Breaking Down the GPT-6 Astra, Sol, and Luna Price Sheet
Rather than forcing engineering teams to choose between an expensive flagship model and older legacy weights, OpenAI now segments text and reasoning inference into three distinct performance bands measured per million tokens:
| Model | Standard Input / 1M | Standard Output / 1M | Cached Input / 1M | Primary Workload | | :--- | :--- | :--- | :--- | :--- | | GPT-6 Astra | $10.00 | $50.00 | $1.00 | Complex reasoning, multi-file coding, planning | | GPT-6.1 Sol | $2.00 | $10.00 | $0.10 | Production coding and agentic tool loops | | GPT-6 Luna | $0.10 | $0.50 | $0.01 | High-volume classification, routing, extraction |
For long-context jobs exceeding 272,000 tokens in a single prompt, standard input and output rates double on input and rise 50% on output across all three models. For instance, GPT-6 Astra scales to $20.00 input and $75.00 output above the 272K boundary, while GPT-6.1 Sol moves to $4.00 input and $15.00 output.
Caching, Batch API, and Regional Residency Economics
Two pricing levers have become central for teams running autonomous workflows at scale. First, automatic prompt caching trims repeated system prompts, tool schemas, and shared repository context by 90% to 95%, turning multi-turn agent loops on GPT-6.1 Sol from a major cloud line item into a predictable expense. Second, asynchronous workloads submitted through the Batch API with a 24-hour completion window receive a flat 50% discount on both input and output rates across every GPT-6 tier.
At the same time, OpenAI has introduced explicit surcharges for speed and compliance. Developers opting into Fast Mode or Ultrafast Mode for latency-sensitive interactive applications pay priority rates above standard tiers, while regulated enterprise and public-sector workloads requiring regional data residency or FedRAMP compliance carry a 10% regional-processing uplift. Specialized vertical endpoints sit higher still, led by GPT-6 Rosalind for life sciences at $15.00 input and $75.00 output per million tokens.
Why AWS Marketplace Procurement Matters for Enterprise Adoption
On the commercial side, listing ChatGPT Enterprise directly in the AWS Marketplace removes a recurring bottleneck for corporate buyers. Large enterprises often hold multi-year committed spend agreements with AWS under Enterprise Discount Programmes. Allowing IT and security leaders to draw down existing AWS cloud commitments to purchase ChatGPT Enterprise seats shortens legal review cycles and consolidates vendor billing.
OpenAI continues to enforce a strict boundary between workspace subscriptions and metered API usage, however. Paying for ChatGPT Enterprise, Business, or Edu seats does not fund or pool credits for developer API calls, which remain billed separately through the OpenAI platform. That separation ensures high-volume automated software cannot inadvertently exhaust a company's employee seat budget, even as both products increasingly target the same enterprise procurement teams.
Competitive Pressure in the Mid-Tier Agent Market
The aggressive $2.00 input and $10.00 output pricing on GPT-6.1 Sol underscores how fierce competition has become in the middle of the model stack. With Anthropic adjusting Claude Fable 5.2 and Mythos 5.2 pricing ahead of its Nasdaq roadshow, Meta pushing commercial 1M-context APIs for Muse Spark 1.1, and Google routing multi-model coding tasks across Gemini 4 Hydrogen, enterprises rarely commit 100% of token volume to a single flagship endpoint.
Instead, production engineering teams are adopting tiered routing: sending classification and log triage to $0.10 models such as GPT-6 Luna, delegating day-to-day coding and tool execution to GPT-6.1 Sol, and reserving GPT-6 Astra strictly for architectural reviews or hard reasoning failures. OpenAI's November price sheet formalises that split as the standard operating pattern for enterprise AI.
For developer teams budgeting 2027 deployments, the spread between tiers is stark. Routing a 10-million-token daily document-extraction job to GPT-6 Luna instead of GPT-6 Astra cuts standard input costs by 99%, and combining cached system prompts with Batch API scheduling lowers batch ingestion outlays even further. Meanwhile, OpenAI has paired this developer segmentation with an infrastructure policy pledge published on 5 November, committing to absorb full electricity generation costs and disclose site-level water and energy figures for new compute campuses without relying on municipal non-disclosure agreements.
Sources
- 1OpenAI API Pricing Documentationdevelopers.openai.com
- 2OpenAI: Introducing GPT-6.1 Solopenai.com
- 3OpenAI Newsroomopenai.com