Anthropic and OpenAI Trade Blows With Dueling AI Model Launches

Two of the world's leading AI labs went head-to-head on September 22, in one of the fastest-turnaround product battles the industry has seen. Anthropic released Claude Opus 5.5 in the morning, touting major gains in coding, research, and knowledge-work performance. Within minutes, OpenAI answered with not one but two new models, GPT-6 Sol and GPT-6 Luna, both priced well below their predecessors. The back-to-back launches underscore how quickly the AI frontier is now moving, and how central price has become to the competition.

What Anthropic shipped

Claude Opus 5.5 is pitched as a faster, more capable successor to Opus 5, with Anthropic citing a roughly 30% improvement in output generation speed. On agentic coding benchmarks, the company reported Opus 5.5 solving 66.4% of tasks on an internal Terminal-Bench 4.0 suite, compared with 55.8% for its Fable 5.1 model. On a science-focused benchmark, Opus 5.5 reportedly nearly doubled the score of the outgoing Opus 5, and on a knowledge-work evaluation it posted a notably higher Elo-style rating.

Anthropic also highlighted safety work alongside the capability jump, saying the new model shows a sharp drop in attempts to circumvent its operating boundaries compared with Opus 5. On pricing, Opus 5.5 lands at $4 per million input tokens and $20 per million output tokens in its standard configuration, with a faster premium tier priced higher and a steep discount on cached inputs.

OpenAI's near-instant response

OpenAI didn't wait to see how the market reacted. Its two new models, Sol and Luna, arrived priced roughly half of what the previous GPT-5.6 generation charged. Sol, the more capable of the pair, is priced at $2 per million input tokens and $10 per million output tokens, while Luna is aimed squarely at high-volume, lower-cost use cases at a fraction of a cent per thousand tokens. On a widely used software-engineering benchmark, OpenAI reported Sol scoring in the high 60% range, with Luna close behind despite its lower price point.

Why the timing matters

Simultaneous or near-simultaneous launches aren't new in AI, but the speed of OpenAI's countermove — reportedly landing within minutes of Anthropic's announcement — suggests both companies are now watching each other's release calendars closely, if not coordinating around anticipated announcement windows. Enterprise customers appear to be noticing the practical impact: a product executive at GitHub said Opus 5.5 resolved more terminal-based tasks in fewer steps than its predecessor, while an AI products lead at Box reported responses running roughly 40% less verbose, a detail that matters for companies paying by the token.

The bigger picture: an AI price war

What's notable across both releases isn't just raw capability — it's that both labs chose to compete on cost at the same time they compete on benchmarks. Just a year or two ago, frontier model pricing tended to climb alongside capability. Now, cheaper tiers like Luna and Anthropic's discounted cache pricing suggest labs are racing to make advanced AI affordable enough for high-volume enterprise workloads, not just premium research use cases. For businesses building on these platforms, the practical upshot is that the cost of running agentic coding, research, and knowledge-work tasks at scale is dropping fast, even as raw performance keeps climbing.

With both labs now shipping faster, cheaper, and more capable models within the same 24-hour window, the pattern looks likely to repeat: expect the next major lab announcement to land with similarly little warning.