AI Price War Erupts as OpenAI, Anthropic Launch New Models
OpenAI's GPT-6 Sol and Luna follow Claude Opus 5.5, cutting top-tier AI costs in half.
The price of top-tier AI intelligence just got cut in half. OpenAI released GPT-6 Sol and GPT-6 Luna on Monday, hours after Anthropic launched Claude Opus 5.5. The most significant detail is not the new models themselves, but their cost. GPT-6 Sol and Luna are priced at 50% of their GPT-5.6 predecessors. Anthropic simultaneously introduced a 20% price cut for its own new flagship model.
This creates a direct, aggressive pricing clash. GPT-6 Sol now matches the price of Grok 4.7 for input tokens and undercuts it on output. GPT-6 Luna becomes one of the cheapest capable models on the market at $0.10 per million input tokens. The cost floor for high-performance AI is dropping fast.
The pricing landscape as of September 22, 2026:
| Model | Input | Cached Input | Output |
|---|---|---|---|
| GPT-6 Luna | $0.10/M | $0.01/M | $0.50/M |
| GPT-5.6 Luna | $0.20/M | $0.02/M | $1.20/M |
| Grok 4.7 | $2/M | $0.50/M | $6/M |
| GPT-6 Sol | $2/M | $0.20/M | $10/M |
| Claude Opus 5.5 | $4/M | $0.20/M | $20/M |
| GPT-5.6 Sol | $4/M | $0.40/M | $20/M |
GPT-5.6 models are also scheduled for a 25% price increase in November, making the new GPT-6 pricing even more stark in comparison. The competitive pressure is immense. Grok 4.7 launched with aggressive pricing to undercut the previous GPT-5.6 line. Now it faces a direct price match from GPT-6 Sol.
OpenAI frames the cuts as a pass-through of infrastructure savings. The company blog states improvements in caching and inference allow it to serve these models at lower cost. The technical claim is that Sol and Luna share training methods with the top-tier GPT-6 Astra. This brings some of Astra's advances in coding, factuality, and alignment to more efficient models.
Performance claims support the value proposition. On OpenAI's internal AutomationBench test, GPT-6 Sol at xhigh effort outperforms Claude Opus 5 at max effort for 9% of the cost per task. On factual reliability, Sol makes roughly half as many mistakes as its predecessor. For developers, this means better performance per dollar on complex business workflows.
Claude Opus 5.5 also received a cost reduction. Input prices fell from $5/M to $4/M. More importantly, cached input prices dropped by 60%, from $0.50/M to $0.20/M. This specifically benefits long-running agentic conversations where most input tokens are cached. Anthropic's blog describes the model as a direct response to user feedback about communication style.
Developer and researcher Simon Willison noted that GPT-6 Luna maintains its predecessor's reputation for excellent price-performance. He called GPT-5.6 Luna his favorite model for building applications. The new version cuts that already low price further. His comparison of AI-generated "pelicans" also revealed a stylistic shift, with GPT-6 models producing more muted colors than the bolder GPT-5.6 generation.
The new pricing makes the economics of building AI applications fundamentally different. When the cost of a powerful reasoning model like GPT-6 Sol drops to $2/M input, experimental and high-volume use cases become viable. The barrier to entry for complex automation tasks is lowered.
A key question remains. This rapid price erosion is driven by competition and infrastructure gains. Can this pace of cost reduction sustain itself as models grow more complex? The current war benefits developers, but the long-term economics for providers need to balance performance gains with sustainable margins.
Source: Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war