OpenAI releases GPT-6 Sol and Luna at half the cost
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026, two models trained with methods similar to its flagship GPT-6 Astra but priced at roughly half their GPT-5.6 equivalents. Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50. On AutomationBench, a test of end-to-end business workflows across 47 tools, Sol at maximum effort outperforms Claude Opus 5 at just 9% of its per-task cost. The releases let companies revisit automation workflows previously too expensive to justify at scale.
In short
- OpenAI released GPT-6 Sol and Luna on September 22 at 50% lower prices than their predecessors, with Sol at $2/$10 and Luna at $0.10/$0.50 per million tokens
- On AutomationBench business workflows, Sol outperforms Claude Opus 5 at just 9% of its per-task cost, shifting economics for operations automation
- Both models hallucinate less than predecessors and improve on coding benchmarks, making them suitable for document processing and agentic workflows
- Lower model costs now let companies automate full-scale volumes rather than samples, removing cost bottlenecks from existing operational workflows
- Anthropic released Claude Opus 5.5 the same day with similar cost-and-speed positioning, signaling a shift toward tiered pricing across frontier labs
Two new models bring frontier-level performance to high-volume automation
OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, 2026. According to OpenAI, both models were trained with methods similar to GPT-6 Astra but optimized for speed and cost. GPT-6 Sol targets complex work tasks including coding and business automation. GPT-6 Luna is designed for high-volume, well-defined tasks that run repeatedly.
Pricing cuts are substantial. GPT-6 Sol costs $2 per million input tokens and $10 per million output tokens, down 50% from the previous GPT-5.6 promotional pricing. GPT-6 Luna costs $0.10 and $0.50, also 50% lower. Both models carry a 1.05-million-token context window and support text and image input.
Practical implications for operations work
The release shifts how teams evaluate automation economics. According to benchmark testing by OpenAI, on AutomationBench—which evaluates agents completing workflows across 47 business tools in sales, marketing, operations, support, finance and HR—GPT-6 Sol at maximum effort scores 33.2% at $0.27 per task. This outperforms Claude Opus 5 at its maximum effort setting at just 9% of the per-task cost.
Teams running document classification, content summarization, ticket routing, campaign analysis or agentic workflows can now revisit tasks that were previously too expensive to automate at full scale. A customer support team might now run quality checks on every incoming ticket rather than a sample. An operations team might analyze every account instead of only high-value ones.
However, lower model cost alone does not guarantee a worthwhile automation. Human review, API calls to other systems, storage, logging and failure handling still cost time and money. The stronger business case appears when lower model cost removes a real bottleneck from an already useful workflow.
Competitive positioning and what to expect next
Anthropic released Claude Opus 5.5 on the same day as OpenAI's announcement, with similar positioning: same training methods as the flagship, optimized for cost and speed. On Artificial Analysis benchmarks, GPT-6 Sol and Luna both show improvements in hallucination rates. According to the hallucination benchmark AA-Omniscience, Sol reduces its hallucination rate from 92% to 60% and Luna from 93% to 77%.
The releases reflect a shift in frontier model economics. For the past year, pricing followed a simple pattern: better capability meant higher cost. GPT-6 Sol and Luna demonstrate that capability and cost can move independently. This likely signals that other labs will continue releasing lower-cost tiers built on existing architectures.
Teams should begin auditing their existing workflows to identify which could now run at scale. A 50% cost reduction on high-volume tasks can change which automation projects clear a business case, especially for mid-size companies processing thousands of documents, tickets or customer interactions monthly.
How AiStaffo would automate this
AiStaffo automates back-office staff work using AI: data entry, follow-ups, reconciliations, reports, billing and document handling. The release of GPT-6 Sol and Luna at half the cost makes it practical to run document classification, ticket routing, invoice processing and account reconciliation on every single item in a queue, not just high-value ones. AiStaffo connects these models to your business systems—your CRM, accounting software, ticketing platform—to run automations that previously required sample-based review or human-touch checkpoints. Sol handles complex multi-step workflows like invoice validation or complaint triage; Luna processes high-volume repetitive work like field extraction or status updates. The owner still reviews exceptions and sets policy; everything else runs automatically. Book a free automation audit to identify which staff workflows AiStaffo can transfer to these new models.
Questions people ask
How much cheaper are GPT-6 Sol and Luna than the previous versions?
Which model should we use for our automation workflows?
How does Sol compare to Claude models?
Do these models work with the tools our team already uses?
When should a company switch to these models?
Book a free automation audit
Thirty minutes. We look at one process you run every week and tell you exactly what an AI worker would take off your desk, and what it would not.
