While the sector debated slowing down progress, both companies are presenting new versions focused on efficiency, lower costs, and better performance for complex tasks
At a time when the artificial intelligence industry is discussing the need to slow down the pace of development, Anthropic and OpenAI decided to move forward with the launch of new models. Both companies are updating their previous systems with versions that maintain similar performance levels but at a significantly lower cost.
Anthropic introduced Opus 5.5, an evolution aimed at enterprise customers. According to the company, the model excels in complex tasks such as detecting and correcting software inefficiencies, financial analysis, and business work. Internal benchmarks indicate that it outperforms GPT-6 Astra in agentic programming on Terminal-Bench 4.0 and FrontierCode v1.1 (Main), which could attract developers.
Beyond capabilities, Anthropic claims that Opus 5.5 generates texts that are easier to understand and attempts to bypass limits around 85 percent less than previous versions. The price also drops: four dollars for input tokens and twenty for output tokens, compared to five and twenty-five dollars for Opus 5.
OpenAI, for its part, announced GPT-6 Sol and GPT-6 Luna. Both were trained using methods similar to those of GPT-6 Astra and offer improvements in professional work, programming, and computer use. The company highlights that they are up to 50 percent cheaper than the promotional price of their GPT-5.6 equivalents.
Anthropic and OpenAI announce more powerful (and cheaper) AI models - Engadget
Prices and improvements in OpenAI models
In detail, GPT-6 Sol costs two dollars for input tokens and ten for output tokens. Luna, designed for lighter uses, drops to ten cents for input and fifty cents for output. OpenAI claims that Sol makes approximately half the factual errors of its predecessor and that in programming it matches the performance of Fable 5.1 at a lower cost.
The new systems also communicate more clearly. Thanks to improvements in prompt caching, they can reuse more context and respond more quickly. In its own tests, OpenAI asserts that they are better aligned: they lie less about the results of their coding work and more firmly reject attempts to bypass safeguards when given unsafe instructions.
Both Anthropic and OpenAI proposed their own criteria for external evaluators to measure the progress and safety of artificial intelligence development. It remains to be seen if these proposals lead to any industry-wide shared standard.
Availability is already confirmed. Anthropic's Opus 5.5 can be used now through Claude, Amazon Web Services, Google Cloud, and Microsoft Azure. OpenAI's GPT-6 Sol and GPT-6 Luna models are enabled in ChatGPT Work and Codex for Plus, Pro, Business, and Enterprise subscribers. Free and Go users will only have access to GPT-6 Luna in the desktop application.
With these launches, both companies reinforce their commitment to providing more accessible tools without sacrificing capability, in a scenario where the debate about the pace of artificial intelligence advancement remains open.
