{"id":14380,"date":"2026-09-24T06:00:45","date_gmt":"2026-09-24T06:00:45","guid":{"rendered":"https:\/\/krikis.lt\/anthropic-claude-opus-5-5-openai-gpt-6-sol-luna\/"},"modified":"2026-09-24T06:00:45","modified_gmt":"2026-09-24T06:00:45","slug":"anthropic-claude-opus-5-5-openai-gpt-6-sol-luna","status":"publish","type":"post","link":"https:\/\/krikis.lt\/en\/anthropic-claude-opus-5-5-openai-gpt-6-sol-luna\/","title":{"rendered":"Anthropic launches Claude Opus 5.5, OpenAI responds with GPT\u20116 Sol and Luna models"},"content":{"rendered":"

On September 22, 2026, SiliconANGLE News reported that two major AI model developers, Anthropic PBC and OpenAI Group PBC, not only didn\u2019t stop there, but also squeezed prices even further. Anthropic released a new version of Claude Opus 5.5, and a few minutes later, OpenAI announced two new GPT-6 models, Sol and Luna, which cost only half as much as their previous versions.<\/p>\n

Opus 5.5: Performance and Price Reduction<\/h2>\n
\"Security<\/figure>\n

Anthropic has officially announced that Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, which is 40 % less than the previous Opus 5. In addition, cache reads, which account for the majority of the cost of agent and encoding tasks, have been reduced by 60 % to $0.20 per million tokens. The faster service mode costs $8 per million input tokens and $40 per million output tokens, but still remains cheaper than the previous model.<\/p>\n

According to SiliconANGLE, Opus 5.5 matches Fable 5.1 on most tasks, with output generation being 30 % faster than Opus 5, which was released in late July. Benchmarks show Terminal-Bench 4.0 (an agent-based coding test) delivering a score of 66.4 %, compared to 55.8 % for Fable 5.1. AutomationBench task completion jumped to 40 % from 26.9 % for Opus 5. In the scientific realm, Terminal-Bench-Science 0.1 gave Opus 5.5 a score of 58.7 %, compared to 29 % for Opus 5.<\/p>\n

In terms of security, Anthropic claims that Opus 5.5 is the strongest model in automated behavioral auditing to date, with 85% fewer attempts to bypass restrictions than Opus 5. Prompt injection resistance has been improved to the level of Fable 5.1. For high-risk tasks related to cybersecurity and biology, access is restricted or redirected to older models (Opus 4.8 or Opus 5).<\/p>\n

OpenAI GPT\u20116 Sol and Luna: Price Review and Goals<\/h2>\n
\"Business<\/figure>\n

OpenAI also announced two new GPT-6 variants, Sol and Luna, on the same day, which are half the price of the previous GPT-5.6 models. Sol costs $2 per million input tokens and $10 per million output tokens, while Luna costs just $0.10 per input and $0.50 per output. Sol is focused on repetitive coding and agent-based work, while Luna is focused on high-volume, routine tasks such as summarization and data mining.<\/p>\n

OpenAI says Sol completes 33.2 % AutomationBench tasks for 27 cents, while DeepSWE v1.1 (a software engineering test) scores 68.8 %, just 1.1 points lower than the previous Claude Fable 5. Luna scores 66.6 % on the same test. Both models make about half as many errors as the previous model, with Luna matching the GPT-5.6 Sol price point at higher effort settings.<\/p>\n

OpenAI has also reduced the cost of reading cached input tokens to 90 %, claiming that this improves hit rate and allows coding agents to switch reasoning efforts without cache breaks. Models return shorter answers with less jargon and fewer low-value details.<\/p>\n

What does this mean for AI models in Lithuania?<\/h2>\n

Both companies\u201e offerings demonstrate that access to powerful yet cost-effective AI models is expanding in Lithuania. Claude Opus 5.5 is already available via the Claude Developer Platform, AWS, Google Cloud, and Microsoft Azure, while GPT\u20116 Sol and Luna are available via the OpenAI API, ChatGPT Work, Codex, and even a free version of Luna in the desktop application. This allows local startups, research institutions, and enterprises to test advanced models without significant costs.<\/p>\n

Additionally, both Anthropic and OpenAI emphasize security measures: Opus 5.5 has advanced alignment tests, and OpenAI models also show better resistance to prompt injection. This is important when integrating AI solutions into critical business systems or public services.<\/p>\n

Conclusions<\/h2>\n

Anthropic and OpenAI have demonstrated that price cuts don't have to mean performance drops. Claude Opus 5.5 delivers Fable 5.1-level performance at a 40% lower price, while OpenAI GPT\u20116 Sol and Luna are even cheaper options for large-scale tasks. These innovations expand the availability of AI models in Lithuania, enabling local developers and businesses to experiment with cutting-edge technologies without significant investment.<\/p>\n

Sources<\/h2>\n