Spokes.wiki Search About
Blog Posting T2 stable source ↗ source url
Published
Created
Updated
Also known as
Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war
See also
claude-opus-5-5 · gpt-6 · astra · llm-api-pricing · openai-api-pricing-docs · anthropic-model-docs

Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war (Willison)

Simon Willison's note of 22 September 2026 on two launches in one day: Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol and GPT-6 Luna, the two smaller siblings of GPT-6 Astra. Arrived by Telegram on 2026-09-23.

T2: Willison ran his own tests (below) and compared the models' prices side by side. The prices themselves are the vendors'; this page checked each one against the vendor's pricing page on 2026-09-23 (anthropic-model-docs, openai-api-pricing-docs). Two of his side remarks did not match OpenAI's page, and both are recorded below instead of repeated.

The prices he lists (per million tokens: input / cached input / output)

Model Input Cached Output
GPT-6 Luna $0.10 $0.01 $0.50
GPT-6 Sol $2 $0.20 $10
GPT-6 Astra $10 $1 $50
Claude Opus 5.5 $4 $0.20 $20
GPT-5.6 Sol $4 $0.40 $20
Grok 4.7 $2 $0.50 $6
Claude Fable 5.1 $10 $0.25 $50

Opus 5.5 is a 20% cut from Opus 5 ($5/$25) with a 128,000-token output limit. GPT-6 Sol costs half of GPT-5.6 Sol, and Luna sits at the price floor of the whole market.

What he tested himself

  • The pelican test (an SVG of a pelican riding a bicycle, his standing informal check; see astra-pelican-grid). Opus 5.5 at maximum thinking used up its output budget while reasoning, cost about $2.56 per failed attempt, and ran for nearly 20 minutes. He calls "max" "effectively useless" for this job. Fable 5.1 at maximum thinking drew what he calls the best pelican any Anthropic model has made. He also saw GPT-6 drawings use a more muted palette than GPT-5.6's.
  • A real deployment: he moved his Datasette Agent demo to GPT-6 Luna and found it "fast and competent at both SQL queries and building HTML and JavaScript."

He says plainly that it will take a while to judge the new models. The post is a first read.

Two remarks the pricing pages do not support (checked 2026-09-23)

  • "Terra and GPT-6 Sol identically priced." OpenAI's page has GPT-5.6 Terra at $2 / $12 and GPT-6 Sol at $2 / $10. Input is the same; output is not. His conclusion (little reason left to use Terra) still holds, and more strongly, since Sol is the cheaper of the two.
  • "A scheduled 25% increase in November" for GPT-5.6 Sol. OpenAI's page says only that GPT-5.6 Sol's promotional pricing is available "at least through November 21, 2026." That is a floor on how long the discount lasts, not an announced rise.

claude-opus-5-5 · gpt-6 · astra · llm-api-pricing · openai-api-pricing-docs · anthropic-model-docs · simon-willison · synthesis