OpenAI halved prices for the GPT-6 Sol and Luna models; independent analysis does not confirm a performance improvement
OpenAI released the GPT-6 Sol and Luna models on 22 September 2026 at prices 50% lower than GPT-5.6 (Sol 2/10 USD, Luna 0.10/0.50 USD per million tokens). However, an independent analysis by Artificial Analysis shows that actual capabilities remain at the level of their predecessors and have even declined on some tasks.
OpenAI released two new reasoning models in the GPT-6 series – Sol and Luna – on 22 September 2026, with prices 50 % lower than their GPT-5.6 predecessors. GPT-6 Sol now costs 2 dollars per million input tokens and 10 dollars per million output tokens (previously 4/20 dollars), GPT-6 Luna 0.10 dollars for input and 0.50 dollars for output (previously 0.20/1.20 dollars). OpenAI attributes the price reduction to improvements in caching and inference and claims to pass the savings directly on to users. Cached input tokens now offer a 90 % discount, and the Terra model, previously the cheapest in the lineup, has been removed from the portfolio. The models are available in ChatGPT Work and Codex for paid tiers, via the API as gpt-6-sol and gpt-6-luna; users on the Free and Go plans have access to Luna only in the desktop app.
OpenAI bases the positioning of the new models primarily on price/performance relative to the Claude models from Anthropic – for example, on the AutomationBench benchmark, Sol at the highest reasoning level is said to outperform Claude Opus 5 at roughly 9 % of its cost per task. According to The Decoder, however, the selection of benchmarks by OpenAI is selective and omits some standard metrics such as GDPval or Terminal-Bench 4.0.
According to the sources, an independent analysis from Artificial Analysis shows that GPT-6 Sol and Luna have halved the cost per task, but overall capabilities remain at the level of GPT-5.6 – with slight improvements on some tests and declines on others. On the coding agent index, Sol improved by 2 points, while Luna declined by 2 points. Both models recorded a decline on the GDPval-AA v2.1 knowledge work benchmark (Sol by about 100 Elo points, Luna by 75), which the analysis attributes to lower-quality presentation of outputs and incomplete answers. The release took place on the same day that Anthropic reduced prices for its Claude Opus 5.5 model, which, according to the sources, sparked broader price competition between the two companies in the cheaper model segment.
Why it matters
For developers and businesses choosing a model based on price/performance for high-volume or repetitive tasks, it matters that a lower price does not automatically mean better output or output of the same quality – according to independent measurements, quality has instead declined on some types of tasks. This directly affects decisions about migrating between models and whether to rely on benchmarks published by the manufacturer itself.
What was added since the original report
Verified updates
-
Anthropic released Claude Opus 5.5 on the same day (2026-09-23), roughly an hour after OpenAI; Claude Opus 5.5 is 20% cheaper than the previous Opus 5.0 ($4/$20 vs $5/$25); Opus 5.5 improved communication style and token efficiency; The model releases launch a 'price war' between the two largest AI companies; GPT-6 Luna and Sol are part of a broader strategy of price competition, not an isolated release
- Anthropic released Claude Opus 5.5 on the same day (2026-09-23), roughly an hour after OpenAI
- Claude Opus 5.5 is 20% cheaper than the previous Opus 5.0 ($4/$20 vs $5/$25)
- Opus 5.5 improved communication style and token efficiency
- The model releases launch a 'price war' between the two largest AI companies
- GPT-6 Luna and Sol are part of a broader strategy of price competition, not an isolated release
-
GPT-6 Sol and Luna are 50% cheaper than their GPT-5.6 predecessors (Sol $4→$2, Luna $0.20→$0.10 for input); Prompt caching offers a 90% discount on cached tokens; The Terra model (previously the cheapest in the series) is no longer available; The models are available in ChatGPT Work, Codex and the API; OpenAI attributes the price reduction to improvements in caching and inference
- GPT-6 Sol and Luna are 50% cheaper than their GPT-5.6 predecessors (Sol $4→$2, Luna $0.20→$0.10 for input)
- Prompt caching offers a 90% discount on cached tokens
- The Terra model (previously the cheapest in the series) is no longer available
- The models are available in ChatGPT Work, Codex and the API
- OpenAI attributes the price reduction to improvements in caching and inference
Two audiences, two different impacts
What this means
For individuals
Developers working with the API from OpenAI can reduce inference costs by up to half while maintaining the same usage pattern (Sol for more complex tasks, Luna for high-volume simple tasks), but independent tests suggest that a lower price may not mean higher or even the same output quality in all cases.
For a business
Businesses gain a cheaper option for reasoning models for high-volume or repetitive tasks (summaries, data extraction, code), but according to an independent analysis, actual output quality has not improved compared with the previous generation and has instead declined on some knowledge work tasks, so deployment decisions should not rely solely on the manufacturer's claims about price/performance.
ProductivityCheck the original
Event sources
confirmed by 3 independent sources · 4 publishers, 3 independent. We count feeds from the same owner only once.