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.
Short answer: 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.
Ars Technica reported on September 22, 2026 that both Anthropic and OpenAI have unveiled fresh model releases aimed at reducing the price of using advanced AI systems. The announcements come as enterprises weigh the trade-off between cutting-edge performance and budget constraints, prompting a shift toward comparison shopping among frontier models.
Anthropic introduced Opus 5.5, the latest iteration of its flagship workhorse model that handles coding and complex knowledge tasks. The company positioned the update as a more affordable alternative to its previous Opus 5 release while still delivering modest gains in capability. According to Anthropic’s own benchmarks, Opus 5.5 outperforms OpenAI’s GPT-6 Astra in certain coding and knowledge-work scenarios, though the advantage is described as slight. The real emphasis, however, lies in the pricing structure. Input and output tokens for Opus 5.5 are priced at $4 and $20 per million tokens respectively, representing a 20 % reduction compared to Opus 5. Cache reads, which dominate costs in agentic and coding workloads, now cost $0.20 per million tokens, a 60 % drop from the earlier version. The model also generates output more than 30 % faster than Opus 5. Anthropic claims that when factoring in the reduced token consumption typical of everyday use, total savings can approach 40 % for standard workloads. The company noted that Opus 5.5 retains the safety safeguards that apply to its Fable 5.1 model, automatically routing potentially risky requests to an older, more conservative model when needed.
OpenAI’s contribution to the cost-cutting wave arrives in the form of GPT-6 Sol and GPT-6 Luna. These models sit beneath the recently released GPT-6 Astra in the company’s lineup and are marketed as efficient, lower-priced options for everyday tasks. Sol is priced at $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. OpenAI states that both Sol and Luna are roughly half as expensive to run as their predecessors, despite being trained with similar methods to Astra. In benchmark tests the new models occasionally show a few percentage points of improvement over earlier versions on specific tasks, but the primary message from OpenAI is the reduction in operating expense rather than a leap in capability.
The announcements reflect a broader trend in the AI industry where the race for ever larger, more powerful models is being complemented by a focus on affordability and operational efficiency. Enterprises that have begun experimenting with model routers-systems that dynamically select the cheapest adequate model for a given request-are finding that the latest offerings from Anthropic and OpenAI fit well into cost-saving strategies. By lowering token prices and improving speed, the new releases aim to keep frontier models relevant even as open-weight alternatives gain traction.
For developers and AI practitioners, the shift means that decisions about which model to deploy are increasingly driven by predictable pricing and integration ease rather than raw performance alone. The articles suggest that organizations are looking for stable, repeatable outcomes that fit within budget constraints, prompting providers to highlight cost metrics alongside modest capability gains. As the market matures, the ability to balance expense with sufficient performance may become a decisive factor in model selection, shaping how teams build and scale AI-powered applications.
Input tokens $4 per million, output tokens $20 per million, cache reads $0.20 per million tokens. These represent a 20% reduction in token prices and a 60% drop for cache reads compared with Opus 5.
Opus 5.5 cuts token costs by 20% and cache-read costs by 60% versus Opus 5, and Anthropic estimates that typical workloads can see total savings approaching 40% when the lower token consumption is factored in.
Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 per million input tokens and $0.50 per million output tokens, each roughly half the expense of their predecessors.
Enterprises are increasingly choosing models based on predictable pricing and integration ease rather than raw performance, using model routers to pick the cheapest adequate model, prompting providers to highlight cost metrics alongside modest capability gains.
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.
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.
Rabbit has launched a cloud-based AI agent called OS3 that runs on any Windows, macOS, or Linux PC without dedicated hardware, letting users connect up to five devices, choose their own AI models, and access the agent via desktop, Telegram, iMessage, or the legacy R1.
NO COMMENTS YET
Comments are open. Have a thought or a question? Share it below.