New Anthropic, OpenAI models make same promise: A little more for a lot less money

Anthropic and OpenAI are emphasizing cost-efficiency in their latest model lineups, advancing the industry conversation from pure capability toward value per query. Anthropic’s Opus 5.5 is presented as a cheaper alternative to its prior Opus 5, with the company asserting that typical workloads at default settings can be about 40% less expensive because the model both lowers per-token pricing and uses fewer tokens to complete tasks. Anthropic also notes that Opus 5.5 inherits previous safety measures: requests deemed to be in sensitive or high-risk domains can be automatically and transparently redirected to an older model, mirroring protections applied to Fable 5.1. The company highlights Opus 5.5’s competence in areas labeled high risk, such as cybersecurity and biology, which is the rationale given for maintaining conservative routing in those cases. OpenAI’s recent additions — GPT-6 Sol and Luna — are described as incremental, cost-focused updates within the GPT-6 family. OpenAI has used a naming convention to indicate capability and cost tiers: Astra is the most capable and expensive, Sol is positioned as an efficient everyday option, Terra as the generalist, and Luna as the fast, low-cost choice. The firm says Sol and Luna were trained using methods similar to GPT-6 Astra and, depending on benchmarks, offer modest capability improvements over their predecessors while costing roughly half as much to run. Taken together, the releases illustrate a growing market emphasis on tiered offerings that trade off peak performance for lower operational expense. Both vendors continue to provide higher-end models for demanding use cases, while promoting new mid- and lower-tier models intended to reduce costs for common workloads.