OpenAI and Anthropic Cut Prices: GPT-6 and Opus 5.5

OpenAI halves the prices of GPT-6 Sol and Luna, while Anthropic lowers the price of Opus 5.5 and revises the cost of caching. The new tiers reach $0.10/$0.50 for GPT-6 Luna. However, a "max" reasoning test has revealed shortcomings in Opus 5.5 on a simple task, with a significant cost.

Opus 5.5 Fails in "Max" Mode and Reaches 128,000 Tokens

In a test aimed at generating an SVG of a pelican on a bicycle, Claude Opus 5.5, set to "max" reasoning mode, failed to produce a response. The model initially identified the request as a classic test, then spent a long time reasoning before hitting its output limit of 128,000 tokens, without providing a result. A second attempt resulted in the same failure. According to the author, this raises doubts about the utility of "max" mode. These two attempts cost $2.56 and took nearly 20 minutes. In contrast, Fable 5.1 in "max" mode reportedly generated the best pelican SVG among the tested Anthropic models.

Declared Uses: Default in Codex and Datasette Agent

GPT-6 Sol and Claude Opus 5.5 are now used as default models in Codex and Claude Code by the author. The Datasette Agent demonstration on agent.datasette.io has been updated to work with GPT-6 Luna, which has proven to be fast and competent for SQL queries as well as for generating HTML and JavaScript in Datasette applications. A comparison grid has been created to confront the models on this type of test, considered useful for comparing different families of models at various reasoning levels.

GPT-6 Luna at $0.10/$0.50 and Sol at a Significant Decrease

OpenAI has significantly reduced its prices: GPT-6 Luna is offered at $0.10 per million input tokens and $0.50 for output, while GPT-6 Sol sees its price halved compared to GPT-5.6 Sol. A 25% price increase for GPT-5.6 is expected in November. According to the author, GPT-6 is half the price compared to the promotional rates of GPT-5.6, and GPT-5.6 Terra is now priced the same as GPT-6 Sol. On the competition side, Grok 4.7 is listed at $2/$6 per million tokens, which was less than half the price of GPT-5.6 Sol, but now has the same entry price as GPT-6 Sol and is also approaching it in output. GPT-6 Luna is among the most affordable OpenAI models ever launched, with only GPT-4.1 Nano ($0.10/$0.40 in April 2025) and GPT-5 Nano ($0.05/$0.40 in August 2025) preceding it. At a higher level, GPT-6 Astra and Claude Fable 5.1 remain at $10/$50. At the entry level, Haiku 4.5 is priced at $1/$5, while GPT-6 Luna costs a tenth of that at $0.10/$0.50.

Opus 5.5 Drops by 20% and Caching Costs Decrease by 60%

Anthropic has reduced the price of Opus 5.5 to $4 per million input tokens and $20 for output, which is 20% less than the $5/$25 applied to versions 4.5 to 5. The cost of cached reads has decreased by 60%, which has a notable impact for long conversations, where over 90% of input tokens can be charged at this rate, according to the author. The new price of Opus 5.5 was equivalent to that of GPT-5.6 Sol before OpenAI halved Sol's prices. According to Thariq Shihipar, Opus 5.5 addresses criticisms regarding communication style, is cheaper per token than Opus 5.0, aligns its intelligence with Fable 5.1, and operates effectively at all levels of effort. It is also expected to handle Blender better.

Close Releases and Upcoming Models at Anthropic

Anthropic launched Opus 5.5, and about an hour later, OpenAI introduced GPT-6 Sol and GPT-6 Luna, following the releases of Grok 4.7 and MiMo v2.6 Flash/Pro. Anthropic is also planning the upcoming release of Sonnet 5.5 and Haiku 5.5. At the higher tier, GPT-6 Astra and Claude Fable 5.1 are offered at $10/$50, while, according to the author, the price war mainly concerns models positioned just below. A comprehensive evaluation of these new offerings will take time, with the author having previously favored GPT-5.6 Luna for development and now believing there is no longer a reason to use Terra.