Despite ongoing discussions within the artificial intelligence industry regarding the need to pace development, major players Anthropic and OpenAI have moved forward with the release of their latest models. Both companies are unveiling new iterations that aim to deliver performance comparable to their previous releases—Anthropic’s Fable 5.1 and OpenAI’s GPT-6 Astra—while significantly reducing costs.
Anthropic’s new Claude Opus 5.5 is specifically tailored for enterprise customers, emphasizing improved capabilities in complex work, software debugging, and financial analysis. According to company benchmarks, Opus 5.5 outperformed GPT-6 Astra in agentic coding tasks on Terminal-Bench 4.0 and FrontierCode v1.1. Beyond raw performance, Anthropic claims the new model produces more understandable writing and adheres to instructions more closely, attempting to circumvent safety boundaries 85 percent less often than prior versions. The model also features a lower price point, with input tokens costing $4 and output tokens $20, compared to $5 and $25 respectively for the previous Opus 5 model.
OpenAI has introduced GPT-6 Sol and GPT-6 Luna, both designed for efficiency. These models offer improvements in professional work, coding, and computer use at up to 50 percent cheaper rates than the promotional pricing of their GPT-5.6 predecessors. GPT-6 Sol is priced at $2 per input token and $10 per output token, while GPT-6 Luna is available at $0.10 for input and $0.50 for output. OpenAI states that GPT-6 Sol makes approximately half as many factual errors as its predecessor and matches the performance of Fable 5.1 at a lower cost. Additionally, enhancements in prompt caching allow these models to reuse context more effectively, resulting in faster responses.
Both companies have highlighted improved alignment and safety measures. OpenAI reports that its new GPT-6 models are less likely to lie about coding results and more resistant to attempts at bypassing safeguards. Both Anthropic and OpenAI have also proposed criteria for third-party evaluators to assess AI progress and safety, although the potential for these to become industry-wide standards remains uncertain.
Anthropic’s Opus 5.5 is currently available to developers via Claude, Amazon Web Services, Google Cloud, and Microsoft Azure. OpenAI’s new models are accessible in ChatGPT Work and Codex for Plus, Pro, Business, and Enterprise customers, with free users and Go subscribers limited to GPT-6 Luna in the desktop application.
Wait, GPT-6 Luna is only ten cents? I need to test if it still hallucinates as much as the free version used to.
Prices are dropping but capabilities are rising. This speeds up the arms race, not slows it down.