Anthropic has launched Opus 5.5, a new version of its primary model for coding and complex tasks. The company reports that input and output token costs are $4 and $20 per million, respectively, representing a 20 percent reduction compared to Opus 5. Cache read costs are $0.20 per million tokens, down 60 percent, with output generation speeds increasing by over 30 percent.
Anthropic estimates that total savings for typical workloads reach 40 percent due to improved token efficiency. The model includes safety protocols for high-risk domains like cybersecurity and biology, which may trigger automatic routing to older models if requests are flagged. Benchmarks indicate that Opus 5.5 shows modest performance gains over OpenAI's GPT-6 Astra in specific coding and knowledge tasks.
OpenAI has introduced new versions of its GPT-6 models, specifically Sol and Luna, which prioritize efficiency and speed. These models were trained using methods similar to the GPT-6 Astra. OpenAI states that Sol and Luna offer performance improvements of a few percentage points over their predecessors while reducing usage costs by 50 percent.
The API pricing for GPT-6 Sol is set at $2 per 1 million input tokens and $10 per 1 million output tokens. For the Luna model, the rates are $0.10 and $0.50 per 1 million tokens, respectively. These releases reflect a broader industry shift where enterprise customers increasingly prioritize cost-effective and predictable deployments over incremental performance gains.
Both companies are responding to competition from open-weight models and the growing use of model routers by organizations. As businesses focus on operationalizing AI, the demand for expensive frontier models is being balanced against the need for more affordable alternatives. This trend suggests a practical shift in the market toward sustainable, long-term AI integration.
