GPT-6 Sol and Luna: 50% Price Cuts and Benchmark Wins Over Claude Opus 5

Seed story: "Introducing GPT-6 Sol and Luna" (OpenAI) · search original Written from facts verified across 3 report(s) — original explainer, not a copy or translation. Sources at the end.

OpenAI has introduced GPT-6 Sol and GPT-6 Luna, slashing API prices by 50% compared to the previous generation to reshape the cost-performance landscape for production AI applications. With GPT-6 Sol now priced at $2 per million input tokens and GPT-6 Luna at just $0.10, developers can access significantly cheaper inference while reportedly outperforming Claude Opus 5 on key benchmarks at a fraction of the cost.

GPT-6 Sol and Luna Launch Details

OpenAI has expanded the GPT-6 family with the introduction of Sol and Luna, joining the existing Astra model. These new entries are designed to replace the previous GPT-5.6 Sol and GPT-5.6 Luna generations within the ChatGPT interface. This update ensures a seamless transition for users, as the older models will be phased out in favor of the improved GPT-6 architecture.

Availability is immediate and universal. The models are rolling out to all users globally across every ChatGPT tier. This includes:

Free plan users
Plus subscribers
Team and Enterprise accounts

For developers, this means the new models are accessible via the standard API endpoints without requiring separate opt-ins. By integrating Sol and Luna directly into the core product, OpenAI signals that these are the default workhorses for the next cycle, simplifying version management and reducing the friction of adopting the latest capabilities in production workflows.

The 50% Price Reduction Breakdown

The pricing structure for GPT-6 Sol and Luna represents a substantial shift from the promotional rates of their predecessors. Specifically, GPT-6 Sol now costs $2 per million input tokens and $10 per million output tokens. Meanwhile, GPT-6 Luna offers a significantly lower entry point at $0.10 per million input tokens and $0.50 per million output tokens.

These figures reflect a 50% reduction compared to the promotional pricing of GPT-5.6 Sol and Luna. For developers, this tiered approach allows for precise budgeting:

GPT-6 Sol: Suited for complex reasoning tasks where higher output costs are justified.
GPT-6 Luna: Ideal for high-volume, low-complexity workloads due to its minimal per-token expense.

This price drop makes it easier to integrate advanced models into production pipelines without the previous cost barriers.

Benchmark Performance vs. Claude Opus 5

The latest data highlights a significant efficiency gap between the new GPT-6 models and their competitors. When comparing GPT-6 Sol against Claude Opus 5, the former demonstrates superior performance at a fraction of the expense. This is particularly evident in complex, multi-step tasks where cost per completion matters most.

AutomationBench: GPT-6 Sol at xhigh effort outperformed Claude Opus 5 at max effort, achieving this at just 9% of the cost per task.
Agents' Last Exam: GPT-6 Sol at max effort scored 56.4%, exceeding Claude Opus 5’s highest score while costing 60% less per task.

For developers, these metrics suggest that GPT-6 Sol can handle demanding agentic workflows without the premium price tag previously associated with top-tier performance.

Cost-Performance Tradeoffs for Production

The 50% price reduction significantly alters the economics of high-effort agentic workflows. With GPT-6 Sol now priced at $2 per million input tokens and $10 for output, developers can run complex automation tasks at a fraction of previous costs. This shift makes it viable to deploy sophisticated agents in production environments without prohibitive overhead, directly impacting how teams budget for AI-driven operations.

Key economic advantages include:

Reduced Task Costs: GPT-6 Sol at xhigh effort outperformed Claude Opus 5 at max effort on AutomationBench at just 9% of the cost per task.
Efficiency Gains: At max effort, GPT-6 Sol scored 56.4% on Agents' Last Exam, exceeding Claude Opus 5’s highest score at 60% lower cost per task.
Scalability: Lower per-token rates allow for higher volume of agent interactions, enabling more robust testing and iterative development cycles.

For developers, this means agentic systems can be scaled more aggressively. The ability to achieve superior benchmark performance at a lower price point encourages the adoption of high-effort models for critical production tasks, reducing the tradeoff between quality and expenditure.

Safety Classifications and Limitations

OpenAI’s Preparedness Framework assigns a "High capability" classification to GPT-6 Sol and Luna in the Cybersecurity and Biological and Chemical domains. This designation signals that these models possess advanced capabilities in sensitive areas, necessitating rigorous internal controls before public release. For developers, this implies that while the models are powerful, they operate under strict safety guardrails designed to mitigate misuse in critical infrastructure and research contexts.

Conversely, the assessment notes that neither model reaches the "High" threshold in the AI Self-Improvement domain. This distinction is crucial for enterprise architects, as it suggests the models are not currently capable of autonomously rewriting their own core logic or significantly accelerating their own development cycles without human intervention.

Cybersecurity: Classified as High capability
Biological/Chemical: Classified as High capability
AI Self-Improvement: Below the High threshold

These classifications help teams understand the operational boundaries of the new models. While they offer significant performance gains, the lack of high-level self-improvement capabilities means that human oversight remains a critical component of the deployment workflow, particularly when integrating these tools into automated pipelines or complex agentic systems.

Migration Strategy for Developers

Transitioning from GPT-5.6 to GPT-6 requires updating your API endpoints to the new model identifiers. Since GPT-6 Sol and Luna replace their GPT-5.6 counterparts within the ChatGPT interface, ensure your application logic accounts for this swap. For API users, the primary benefit is immediate cost reduction; you should verify that your billing dashboards reflect the new rates, such as $2 per 1 million input tokens for Sol.

To monitor performance changes effectively, implement the following steps:

Update model parameters to target GPT-6 Sol or Luna.
Re-run your existing test suites to validate output consistency.
Track latency and error rates during the initial rollout period.

Because these models are now available across all global ChatGPT plans, you can also test behavior in the consumer interface before full production deployment. This dual approach helps identify any subtle shifts in reasoning or safety responses, ensuring your workflow remains stable while capitalizing on the 50% price cuts.

FAQ

What are the API prices for GPT-6 Sol and GPT-6 Luna?

GPT-6 Sol costs $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. These rates represent a 50% reduction compared to the promotional pricing of the previous GPT-5.6 models.

How do GPT-6 Sol and Luna compare to Claude Opus 5 in terms of performance and cost?

GPT-6 Sol outperformed Claude Opus 5 on AutomationBench at 9% of the cost per task when set to xhigh effort. Additionally, GPT-6 Sol at max effort scored 56.4% on Agents' Last Exam, exceeding Claude Opus 5's highest score at 60% lower cost per task.

Will GPT-6 Sol and Luna be available to all ChatGPT users?

Yes, GPT-6 Sol and GPT-6 Luna are being launched to all users in ChatGPT across both free and paid plans globally. These new models will replace GPT-5.6 Sol and GPT-5.6 Luna within the ChatGPT interface.