TL;DR: On Sept. 22, 2026, Anthropic released Claude Opus 5.5 and OpenAI released GPT-6 Sol and GPT-6 Luna, each promising modest capability gains at sharply lower prices for enterprise and developer use.

Dual launch puts price first

Anthropic and OpenAI each announced new, less costly artificial intelligence models on Tuesday, Sept. 22, 2026, according to CNBC. The timing made the pair the first releases from either lab since fears erupted around AI's potential threat to humanity and since Anthropic CEO Dario Amodei called for an industrywide slowdown, CNBC reported.

Anthropic unveiled Claude Opus 5.5, described as the latest version of its main mass-market workhorse model for tasks like coding and other complex knowledge work, according to Ars Technica. OpenAI introduced GPT-6 Sol and GPT-6 Luna, described as the latest versions of its middle-of-the-road or smaller models focused on efficiency and speed, Ars Technica reported.

The releases were not framed as breakthroughs in new capabilities. Rather, they were about efficiency, Ars Technica reported. As both OpenAI and Anthropic target enterprise customers, they are racing to compete with open-weight models as organizations have explored changing practices and using model routers to use pricey frontier models less in favor of cheaper alternatives, according to that same reporting.

CNBC reported that both frontier labs continue to face stiff competition from cheaper, open-weight competitors. CNBC named Chinese firms Alibaba, Moonshot AI and DeepSeek as sources of pressure, and said the new releases arrived as OpenAI and Anthropic work to satisfy customers who are looking for more cost-effective models and seeking to rein in AI spending.

Engadget described both companies as iterating on their previous releases, Fable 5.1 and GPT-6 Astra, with new models that offer similar levels of performance but at lower costs. The outlet noted that conversation in the AI industry had largely focused on slowing down development or "pacing the frontier," but that did not stop the two launches.

Opus 5.5: flagship positioning and lower tokens

Opus 5.5 sits at the higher end of the models announced that day, but the wider context is that Anthropic is playing catch-up in its race with OpenAI, according to Ars Technica. OpenAI earlier in September released GPT-6 Astra, which has sometimes been modestly beating Opus 5 in benchmarks and user sentiment, Ars Technica reported. By price and capability, Astra is competing with both Opus and Fable, according to that report.

Benchmarks by Anthropic and its partners now show Opus 5.5 performing better at coding and knowledge work than GPT-6 Astra in some cases, albeit modestly, Ars Technica reported. Engadget reported that Anthropic's benchmarks claim Opus 5.5 also scored better at agentic coding than GPT-6 Astra on both Terminal-Bench 4.0 and FrontierCode v1.1 (Main), which could make it more appealing for developers.

Anthropic pitched Opus 5.5 for complex work, "finding and fixing inefficiencies in software" and financial analysis and business work, pitches described by Engadget as laser-targeted at enterprise customers. Engadget also reported that Anthropic claims Opus 5.5 produces writing that is easier to understand, and that it "attempted to circumvent boundaries around 85 percent less often" than past models, suggesting it will disobey directions less often.

Cost is the central change. Anthropic set input and output tokens at $4 and $20 per million, 20% less than Opus 5, according to Ars Technica. Cache reads, which make up the majority of agentic and coding work costs, were priced at $0.20 per million tokens, 60% less than Opus 5, Ars Technica reported. Opus 5.5 also generates output more than 30% faster than Opus 5, according to that announcement coverage.

Anthropic further claims savings closer to 40 percent compared with Opus 5 for typical workloads at default settings, because in addition to lower token prices, Opus 5.5 also uses fewer tokens when completing tasks, Ars Technica reported. CNBC described the latest release as a more token-efficient version that will cost around 40% less to run than Opus 5.

Dianne Penn, head of product management, research and labs at Anthropic, told CNBC in an interview: "One of the things we're continuing to innovate on is how to make that thinking, how to make the answering more efficient, so it uses less tokens depending on your effort setting." Engadget listed the prior Opus 5 prices as $5 input and $25 output, underscoring the size of the listed reduction.

Opus 5.5 is said to be notably capable in "high risk areas" like cybersecurity and biology, so the same protections that applied to Fable 5.1 will also apply here, according to Ars Technica. Under that approach, requests might be automatically and transparently routed to an older model if flagged as treading into protected territory.

GPT-6 Sol and Luna: iterative efficiency tier

OpenAI's release is an iterative step forward, according to Ars Technica. Not long ago, the company introduced its Sol, Terra, Luna naming convention for its GPT-5.6 family of models, and more recently introduced GPT-6 Astra, which has been both its most advanced and most powerful model, Ars Technica reported.

The names carry distinct roles. Astra is the most aggressively powerful and pricey model, meant for heavy-duty coding, research, and similar work, according to Ars Technica. Sol is a capable but more efficient and affordable alternative, meant as a daily driver for such tasks, while Terra is the balanced, general-use model and Luna is the fast, cheap option, Ars Technica reported.

A rough mapping would position Astra against Anthropic's Fable, Sol against Opus, Terra against Sonnet, and Luna against Haiku, but that mapping would be imperfect, especially for higher-end models, Ars Technica reported. CNBC described Sol, a tier below Astra, as meant for more complex workloads like coding, while Luna is geared toward "high-volume tasks" such as extracting information or summarizing documents.

OpenAI says GPT-6 Sol and Luna were trained with similar methods to those used to train GPT-6 Astra, according to Ars Technica. Depending on benchmark, they are sometimes a few percentage points more capable than predecessors at certain tasks, but they cost half as much to use, Ars Technica reported. CNBC reported that OpenAI slashed the API prices by 50% compared with GPT-5.6 promotional pricing. Engadget similarly reported improvements across professional work, coding and computer use, but for up to 50 percent cheaper than the promotional pricing of GPT-5.6 counterparts.

Listed API pricing is $2 per 1 million input tokens and $10 per 1 million output tokens for GPT-6 Sol, according to Ars Technica. For Luna, pricing is $0.10 and $0.50, respectively, Ars Technica reported. Engadget listed the same figures and noted that prompt-caching improvements help the models reuse more context and respond faster.

On quality, OpenAI says both new GPT-6 models are better at getting facts right, with Sol making about half as many mistakes as its predecessor, Engadget reported. In coding, OpenAI says GPT-6 Sol is able to match the performance of Fable 5.1 at a lower cost, Engadget reported. The outlet also reported clearer communication from the new models.

On alignment, OpenAI says the GPT-6 models are more aligned than before based on company tests, Engadget reported. As examples, they lie less about coding work results and deny more attempts to bypass safeguards when receiving an unsafe command, according to that coverage. Like Anthropic, OpenAI has also proposed its own criteria for third-party evaluators to use when judging progress and safety of AI development, Engadget reported.

Availability and enterprise context

Anthropic's Opus 5.5 model is available to developers now through Claude, Amazon Web Services, Google Cloud and Microsoft Azure, Engadget reported. OpenAI's GPT-6 Sol and GPT-6 Luna are both available in ChatGPT Work and Codex for Plus, Pro, Business and Enterprise customers, Engadget reported. Free users and Go subscribers will have access to only GPT-6 Luna in the company's desktop app, according to that same report.

The enterprise focus was explicit across coverage. Ars Technica described developers and staff working on enterprise AI deployments as comparing the two companies' offerings, with cost as a big part of that comparison. Engadget tied Opus 5.5's coding, financial analysis and business-work emphasis directly to enterprise customers.

Slowdown debate forms backdrop

The Sept. 22 launches were the first to roll out since Amodei called for an industrywide slowdown on developing advanced AI as debate about model safety reached a fever pitch, CNBC reported. Former Anthropic researcher Jacob Coxon set off the debate when he posted to X on Sept. 8 that he had quit his job, warning that both labs were "gambling with our lives," CNBC reported.

OpenAI CEO Sam Altman and Tesla and SpaceX CEO Elon Musk also joined Amodei's call, CNBC reported. Engadget noted "a lot of back and forth about slowing the pace of AI development" before the releases.

Ars Technica's analysis argued that discourse around frontier models is chaotic, with social-media claims about one-shot complex 3D video games alongside news of security breaches and other alignment issues, plus calls for slowdowns and regulation. That analysis also said orchestration and harnesses for models are at least as important as the models themselves, naming Anthropic's, OpenAI's, Alibaba's and other large players' systems. It argued some developers and enterprises are less focused on demanding better performance and more focused on predictable deployments and reasonable costs.

Sources referenced