Lower-Cost AI Models
OpenAI Launches Cheaper GPT-6 Sol and Luna Models for Professional Workflows
Anonymous
Key Points
1
OpenAI said the GPT-6 pricing is the standard rate, unlike the GPT-5.6 prices, which had been promotional. The new rates are therefore intended to be the models’ default prices rather than temporary discounts.
2
On Zapier’s AutomationBench, which evaluates business workflows, GPT-6 Luna improved by 5.4 percentage points over its predecessor.
3
On the DeepSWE v1.1 software-engineering benchmark, GPT-6 Sol scored 68.8% at maximum effort, close to Anthropic’s Fable 5 score of 69.9% at its highest effort setting, while OpenAI said Sol operated at about 20% of the cost.
4
OpenAI did not provide a complete set of benchmark results before launch, marking a departure from some of its previous model releases; the published comparisons were therefore limited at the outset.
5
The launch article noted that OpenAI has not announced a GPT-6 Terra model, leaving Astra, Sol and Luna as the identified models in the GPT-6 lineup.
OpenAI launched GPT-6 Sol and Luna on the same day Anthropic released Claude Opus 5.5, intensifying competition in the AI market with dramatic price cuts. Sol and Luna are cheaper versions of the flagship GPT-6 Astra, priced at $2 per million input tokens and $10 per million output tokens for Sol, and just $0.10 and $0.50 for Luna BBN Times. The cuts represent at least 50% savings compared to GPT-5.6 pricing, which OpenAI says reflects genuine cost improvements rather than temporary promotions.
Sol targets complex coding and reasoning tasks, while Luna handles fast, inexpensive work like summarization and data extraction Trading Key. OpenAI claims the new models deliver greater accuracy and fewer coding errors, though complete benchmark results were not released at launch The Neuron Daily. The rollout highlights OpenAI's push to make advanced AI affordable for professional workflows and automation.
Sol and Luna Fill the Middle Ground Between Astra and Cost
GPT-6 Sol is built for coding, reasoning, and complex problem-solving. It costs far less than Astra but more than Luna. On DeepSWE v1.1, a software-engineering benchmark, Sol scored 68.8% at maximum effort Archy Netys, nearly matching Anthropic's Fable 5 at 69.9% while operating at about 20% of the cost. Luna is the budget option, designed for high-volume tasks that don't require deep reasoning.
On Zapier's AutomationBench, which tests business workflows, Luna improved 5.4 percentage points over its predecessor Trading Key. OpenAI stressed these are permanent prices, not promotions. The company attributes the savings to better caching and faster inference, allowing costs to drop without sacrificing quality.
Anthropic Fires Back Hours Later With Opus 5.5 Price Drop
Anthropic's same-day release of Claude Opus 5.5 with its own price cuts directly countered OpenAI's competitive advantage The Neuron Daily. The timing neutralized comparisons OpenAI had planned. Both companies are locked in a pricing war as they race to make enterprise AI affordable while maintaining profit margins. Neither firm disclosed complete benchmarks at launch, leaving performance claims partially unverified.
The competitive pressure underscores how AI pricing has become a key battleground. OpenAI's move to establish these as default rates rather than promotions signals confidence in its cost structure Digg. Yet without full benchmark transparency, customers must evaluate models partly through word-of-mouth and limited public data.
Missing Benchmarks and Concerns About AI Agent Oversight
OpenAI departed from prior practice by not releasing complete benchmark results before launch BBN Times. This limits independent verification of performance claims and makes side-by-side comparison difficult for buyers. The company disclosed only select benchmarks where Sol and Luna performed competitively, raising questions about broader performance across other tasks.
The launch occurs as AI safety experts warn about increasingly capable agents interacting with computers and external systems Archy Netys. Stronger human oversight and safety mechanisms are seen as critical before deploying these models at scale. OpenAI has also not announced a GPT-6 Terra model, leaving Sol, Luna, and Astra as the identified entries in the GPT-6 family for now.
Sentiment
Mostly NegativeNEGATIVENEUTRALPOSITIVE
Publishers
77
Articles
122
Reach
199
A
AnonymousAuthor
