The industry's been talking a lot lately about "pacing the frontier"—slowing down the advancement of frontier models. But on September 22, 2026, Anthropic and OpenAI announced new models almost simultaneously, proving that at least on the pricing front, neither side has any intention of slowing down.
Anthropic's new Claude Opus 5.5 targets scenarios enterprise customers care about: identifying and fixing software inefficiencies, financial analysis, and business document work. According to Anthropic's own benchmarks, Opus 5.5 outperforms OpenAI's GPT-6 Astra in agentic coding on both Terminal-Bench 4.0 and FrontierCode v1.1 (Main)—a selling point for developers. On pricing, Opus 5.5's input tokens dropped to $4 per million and output to $20, a notable cut from the previous Opus 5's $5 input and $25 output. Anthropic also stated that the new model "attempts to circumvent safety boundaries roughly 85% less often" than before, with writing output that's also easier to read.
OpenAI, meanwhile, launched two models at once: GPT-6 Sol and GPT-6 Luna, both trained using the same methodology as GPT-6 Astra. They focus on improvements to professional work, coding, and computer-use capabilities, priced up to 50% cheaper than GPT-5.6's promotional pricing. GPT-6 Sol costs $2 per million input tokens and $10 for output; Luna is even cheaper at $0.1 input and $0.5 output. OpenAI says Sol's error rate is roughly half that of its predecessor, with coding performance matching Anthropic's Fable 5.1—but at a lower cost. Combined with improvements to prompt caching, both new models can reuse more context and respond faster.
On alignment testing, OpenAI says its internal tests show the GPT-6 series lies less often when reporting code execution results, and shows a higher rate of refusing to bypass safety mechanisms when given unsafe instructions. Like Anthropic, OpenAI has also proposed a set of evaluation criteria for third-party assessors to use in measuring AI progress and safety. Whether these proposals will become industry-wide standards remains to be seen.
As for availability, Opus 5.5 is now open to developers via Claude, Amazon Web Services, Google Cloud, and Microsoft Azure. GPT-6 Sol and GPT-6 Luna have landed in ChatGPT Work and Codex, available to Plus, Pro, Business, and Enterprise users; free-tier users and Go subscribers currently only have access to GPT-6 Luna through the desktop app.
