Anthropic and OpenAI each shipped cheaper frontier models on Sept. 22, cutting the cost of running their latest systems as customers push to rein in AI spending.
Anthropic unveiled Claude Opus 5.5. The company said the model matches Claude Fable 5.1 on most work while costing about 40% less to run than Opus 5 on typical workloads, combining lower per-token prices with fewer tokens per task. List prices are $4 per million input tokens and $20 per million output tokens—20% below Opus 5—with cache reads at $0.20 per million, a 60% cut that matters for long agent sessions, according to Anthropic. The company also said Opus 5.5 was tested before release by independent evaluators including Frontier Design and METR.
OpenAI introduced GPT-6 Sol and GPT-6 Luna, mid-tier GPT-6 models below GPT-6 Astra. The company cut API prices for both by 50% versus GPT-5.6 promotional rates: Sol to $2/$10 per million input/output tokens, and Luna to $0.10/$0.50. OpenAI pitched Sol for complex work such as coding and Luna for high-volume tasks like extraction and summarization. The models are available in ChatGPT Work and Codex for paid tiers and via the API as gpt-6-sol and gpt-6-luna, with a gradual ChatGPT rollout, the company said.
Contemporaneous coverage put the launches roughly 90 minutes apart—with some reports placing Anthropic around 9:30 a.m. PT and OpenAI around 11:00 a.m. PT—though neither company published exact public timestamps in their product posts. CNBC framed the dual rollout as the first major releases from either lab since Anthropic CEO Dario Amodei called for pacing the frontier; that “first major releases” characterization is CNBC’s framing, not an independently established industry fact. CNBC also reported that OpenAI CEO Sam Altman and Tesla/SpaceX CEO Elon Musk had joined Amodei’s pacing rhetoric after a former Anthropic researcher’s Sept. 8 warning inflamed the safety debate.
Both companies cast the cuts as efficiency wins—better caching and cheaper inference. CNBC pointed to pressure from cheaper open-weight rivals, including Chinese labs such as Alibaba, Moonshot AI, and DeepSeek. OpenAI reported that on DeepSWE, a software-engineering benchmark, GPT-6 Sol at max effort scored within about a point of Anthropic’s larger Fable 5 at roughly 80% lower cost per task; those figures are company-reported and have not been independently audited for this story. Anthropic said Opus 5.5 beats prior Claude tiers on several coding and knowledge-work benchmarks while undercutting Opus 5 on cost per task—again, company metrics.
Enterprises shopping cost-per-task will likely reprice agent workloads after twin same-day price moves. Whether safety rhetoric outlasts another commercial race is a question for the next cycle, not this price sheet.