Anthropic and OpenAI roll out new AI models: modest gains, much lower prices

Anthropic and OpenAI are both refreshing their AI model lineups, and the latest releases point to a shared pattern: modest capability gains paired with meaningfully lower prices.

Anthropic says its Opus 5.5 model delivers roughly 40 percent savings compared to Opus 5 for typical workloads run at default settings. Part of that saving comes from lower per-token pricing, but Opus 5.5 also completes tasks using fewer tokens overall. The company notes that Opus 5.5 shows notable strength in "high risk" domains such as cybersecurity and biology. Because of this, the same safeguards already applied to the Fable 5.1 model will extend to Opus 5.5 — requests flagged as touching sensitive territory may be automatically and invisibly rerouted to an older model.

Sol and Luna: a cost-focused update from OpenAI

OpenAI has released refreshed versions of its GPT-6 Sol and Luna models, part of a broader naming system the company introduced for its GPT-5.6 family: Sol, Terra, and Luna. Just this month, OpenAI also introduced GPT-6 Astra, currently its most advanced and powerful model.

Under this naming scheme, Astra targets demanding tasks like heavy coding and research and comes at the highest price point. Sol serves as a more efficient, affordable option for everyday use, Terra is positioned as the balanced, general-purpose model, and Luna is the fastest and cheapest tier. These roughly correspond to Anthropic's lineup — Astra to Fable, Sol to Opus, Terra to Sonnet, and Luna to Haiku — though the comparison is imperfect, especially at the higher end.

OpenAI says GPT-6 Sol and Luna were trained using methods similar to those behind GPT-6 Astra. Depending on the benchmark, they show modest improvements over their predecessors while costing half as much to use.