Anthropic’s Opus 5.5 and OpenAI’s GPT-6 Sol and Luna are the new cost-focused models released for enterprise users. Opus 5.5 cuts input token costs to $4 per million and output to $20 per million, while Sol charges $2 input and $10 output, and Luna just $0.10 input and $0.50 output per million tokens.
Short answer: Anthropic’s Opus 5.5 and OpenAI’s GPT-6 Sol and Luna are the new cost-focused models released for enterprise users. Opus 5.5 cuts input token costs to $4 per million and output to $20 per million, while Sol charges $2 input and $10 output, and Luna just $0.10 input and $0.50 output per million tokens.
On September 22, 2026, Ars Technica reported that both Anthropic and OpenAI have introduced new model families that emphasize lower prices while offering modest performance gains. The announcements come as the competition among frontier AI providers shifts from raw capability to price efficiency, prompting developers and enterprise teams to weigh options more carefully.
Anthropic’s latest release is Opus 5.5, positioned as the updated version of its primary workhorse model used for coding and complex knowledge tasks. The company says the new version cuts input token costs to four dollars per million and output token costs to twenty dollars per million, which represents a twenty percent reduction compared to the previous Opus 5. Cache reads, which dominate expenses in agentic and coding workloads, are now priced at twenty cents per million tokens, a sixty percent drop. Opus 5.5 also generates output more than thirty percent faster than its predecessor. Anthropic estimates that typical workloads see savings closer to forty percent because the model not only costs less per token but also tends to consume fewer tokens overall. The firm notes that Opus 5.5 retains strong abilities in high-risk domains such as cybersecurity and biology, and that the same safety routing used for earlier models will automatically divert flagged requests to an older, more conservative version.
OpenAI’s update arrives in the form of GPT-6 Sol and Luna, two models that sit below the flagship GPT-6 Astra in the company’s lineup. Sol is marketed as a capable yet efficient daily driver for tasks like coding and research, while Luna is positioned as the fastest and cheapest option. OpenAI states that both models were trained using methods similar to those employed for Astra, and that depending on the benchmark they show only a few percentage points of improvement over their predecessors. The main advantage, however, is cost: Sol’s API charges two dollars per million input tokens and ten dollars per million output tokens, whereas Luna costs just ten cents per million input tokens and fifty cents per million output tokens. These prices are roughly half what users paid for the earlier generation of mid-tier models.
The pricing moves reflect a broader trend in which enterprise customers are increasingly interested in predictable, affordable deployments rather than chasing the absolute top of the performance curve. Organizations have begun experimenting with model routers that direct routine work to cheaper, open-weight alternatives while reserving the most expensive frontier models for the most demanding cases. Both Anthropic and OpenAI argue that their new releases push the frontier forward in modest ways while delivering substantial cost reductions, thereby addressing the desire for efficient, production-ready AI.
Safety considerations remain part of the conversation. Anthropic highlights that Opus 5.5 inherits the protective mechanisms of its earlier Fable 5.1 model, meaning that requests that touch on sensitive topics such as cybersecurity or biology may be silently rerouted to a more cautious version. OpenAI has not detailed comparable safeguards for Sol and Luna, but the company emphasizes that the models have undergone the same alignment training as their more powerful siblings.
Beyond raw specifications, the article notes a growing recognition that the models themselves are only one piece of the AI puzzle. Orchestration layers, runtime harnesses, and organizational practices are now seen as at least as important for achieving reliable results. As a result, some developers and enterprise leaders are shifting focus from demanding ever-greater performance to seeking stable, cost-effective implementations that can be maintained over time. This shift could act as a natural slowdown in the relentless push for larger models, with the newly announced offerings tailored to that emerging priority.
Anthropic's latest release is Opus 5.5, positioned as the updated version of its primary workhorse model used for coding and complex knowledge tasks.
Opus 5.5 charges $4 per million input tokens and $20 per million output tokens, a 20% reduction from Opus 5; cache reads cost $0.20 per million tokens, down 60%.
OpenAI released GPT-6 Sol and Luna; Sol costs $2 per million input tokens and $10 per million output tokens, while Luna costs $0.10 per million input tokens and $0.50 per million output tokens.
Anthropic says Opus 5.5 retains the safety routing of earlier models, automatically diverting flagged requests on sensitive topics like cybersecurity or biology to an older, more conservative version.
Source: Ars Technica
UK AI Security Institute and the EvalEval Coalition have published reproducible benchmark results, releasing Evaluation Cards for five core benchmarks-HealthBench, FrontierMath, Humanity's Last Exam, SWE-Bench Pro and Terminal-Bench 2.0-tested on six frontier LLMs and two cyber-focused evaluations, using the Every Eval Ever schema to ensure transparency.
Anthropic’s Opus 5.5 and OpenAI’s GPT-6 Sol and Luna models lower token prices-Opus 5.5 at $4 input and $20 output per million tokens (20 % cheaper than Opus 5) and Sol/Luna at $2/$10 and $0.10/$0.50 per million tokens, roughly half the cost of their predecessors-offering developers reduced AI expenses.
Microsoft announced on September 22, 2026 that it helped dismantle the subscription-based AI-driven fraud platform EvilTokens, which had compromised roughly 12,000 Microsoft accounts across about 10,000 organizations worldwide after appearing on Telegram in February 2026.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.