OpenAI Launches GPT-6 Sol and Luna: Halving API Costs with 1.05M Context Window
OpenAI has released GPT-6 Sol and Luna, cutting API pricing by 50% versus GPT-5.6 with a 1.05M token context window, automatic prompt caching, and expanded desk
On September 22, 2026, OpenAI officially announced the release of GPT-6 Sol and GPT-6 Luna, expanding the GPT-6 model family alongside its premier flagship, GPT-6 Astra. Available immediately across the OpenAI API and ChatGPT tiers, both models establish an aggressive 50% token price cut compared to promotional rates for the previous GPT-5.6 generation, drastically lowering the operating overhead for multi-step agentic systems and high-volume enterprise traffic.
Image source: OpenAI
The launch marks a transition from a single-flagship posture toward a structured model hierarchy. While GPT-6 Astra handles frontier reasoning, Sol and Luna target the practical friction points of production software: context size, inference latency, and API unit economics.
A 50% API Cost Reduction: Pricing and Core Architecture
The centerpiece of the Sol and Luna rollout is an aggressive pricing restructure that halves token rates relative to GPT-5.6 promotional levels.
- GPT-6 Sol Pricing: Priced at $2.00 per million input tokens and $10.00 per million output tokens, exactly half of GPT-5.6's previous rates of $4.00 input and $20.00 output per million tokens.
- GPT-6 Luna Pricing: Priced at $0.10 per million input tokens and $0.50 per million output tokens, down sharply from the previous lightweight tier rates of $0.20 input and $1.20 output per million tokens.
- Batch API Discounts: Asynchronous workloads that do not require immediate interactive latency qualify for OpenAI's Batch API, granting an additional 50% discount on top of the already reduced base rates.
- 1.05M Token Context Window: Both Sol and Luna natively feature a 1,050,000 token context window. This capacity allows engineering teams to ingest full software repositories, voluminous regulatory filings, or dozens of conversational history logs in a single call without manual truncation.
- Automatic Prompt Caching: The updated runtime automatically recognizes and caches repeated prompt prefixes, such as system prompts, documentation corpora, and structured tool definitions, reducing both latency and recurring input costs.
Historically, million-token context windows came at premium price points that limited continuous real-world use. The combination of lower unit costs and automatic caching makes million-token context feasible for production workflows.
Workload Specialization: Agentic Coding vs. High-Throughput Routing
Rather than presenting generalized benchmarks, OpenAI has segmented Sol and Luna into two complementary operational roles.
GPT-6 Sol: Daily Engineering and Multi-Step Agents
GPT-6 Sol is built for everyday software engineering, complex multi-turn reasoning loops, and core enterprise automation.
According to OpenAI's internal evaluation data, GPT-6 Sol achieves a 50% reduction in error rates compared to previous-generation models when executing multi-step agentic tasks. The model demonstrates improved instruction retention and tool-calling reliability across extended execution chains. However, independent evaluations by analysis groups including Artificial Analysis note that Sol's raw benchmark scores align closely with GPT-5.6 performance. The fundamental advance in Sol lies in execution consistency, reduced flakiness, and halved operating cost, rather than fundamental frontier reasoning breakthroughs, which remain exclusive to Astra.
GPT-6 Luna: High-Throughput Ingestion and Routing
GPT-6 Luna operates as an ultra-fast gateway layer where low latency and massive query volume dominate.
- Intent classification and dynamic dispatch to specialized models or external APIs
- High-speed extraction of structured JSON entities from unstructured text
- Real-time content moderation, sanitization, and input preprocessing
At $0.10 per million input tokens, developers can place Luna as a ubiquitous ingress router in front of user-facing systems, filtering queries and triggering downstream tools without inflating infrastructure budgets.
Availability Across OpenAI API, ChatGPT, GitHub Copilot, and Azure
OpenAI deployed GPT-6 Sol and Luna across consumer, developer, and cloud distribution channels on day one.
- OpenAI API: Developers can call both models immediately using the identifiers
gpt-6-solandgpt-6-luna. - ChatGPT and Codex: Both models are active for paid subscribers across Plus, Pro, Business, Enterprise, and Edu plans, as well as within ChatGPT Work and the Codex development harness.
- Desktop Application Access: Free and Go tier users can access GPT-6 Luna directly within OpenAI's official desktop application for daily assistive tasks.
- Ecosystem Integration: Microsoft Foundry (Azure) and GitHub Copilot have enabled native support for both Sol and Luna from launch date.
This release signals a pragmatic maturation of OpenAI's portfolio: reserving Astra for complex problem-solving while deploying Sol and Luna to power high-frequency agentic software at scale.