Opus 5.5 and GPT-6 Sol/Luna cut serving prices

Opening

Anthropic and OpenAI each shipped cheaper mid/high tiers Tuesday — list prices down on the order of 40–50% versus the prior generation, framed as serving-cost pass-through. Anthropic’s Opus 5.5 post (first model in the Claude 5.5 family) says the model “performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5,” with input/output at $4 / $20 per million tokens (vs $5 / $25 for Opus 5) and cache reads at $0.20 (vs $0.50). Anthropic says Opus 5.5 was external-tested before release by Frontier Design and METR, and is its first release since Amodei’s pacing call. Separately, OpenAI’s Sol/Luna post cuts Sol and Luna API prices 50% versus GPT-5.6 promotional pricing: Sol $4→$2 input / $20→$10 output; Luna $0.20→$0.10 / $1.20→$0.50 per million tokens. Astra remains the flagship; Sol/Luna available in ChatGPT Work and Codex for paid tiers (Luna also Free/Go desktop); not yet in Chat. CNBC (CJ Haddad) attributes the Anthropic “~40% less to run” line to Dianne Penn and frames both releases as the first since the recent slowdown debate.

The numbers

Opus 5.5 list — $4 / $20 per 1M in/out; cache reads $0.20 — Anthropic primary (vs Opus 5 $5 / $25 / $0.50).

Typical-workload claim — ~40% less cost vs Opus 5 at default settings — Anthropic (company test; also cites fewer tokens per task).

GPT-6 Sol — $2 / $10 per 1M in/out (50% vs GPT-5.6 Sol promo $4 / $20) — OpenAI primary.

GPT-6 Luna — $0.10 / $0.50 per 1M in/out (50% vs GPT-5.6 Luna promo $0.20 / $1.20) — OpenAI primary.

Fast mode Opus 5.5 — $8 / $40 per 1M in/out, up to 2.5× speed — Anthropic primary.

Why it matters

Both labs are cutting list serving prices while claiming higher or matched capability on agentic/coding workloads — the same mechanism the inference-margin theme tracks (cost per token falling at least as fast as price). Anthropic’s external pre-release eval footnote (METR / Frontier Design) sits beside Friday’s Accenture embedded-evaluator structure; it does not replace it. Competitive bench tables in both posts are vendor-selected and should not be treated as independent scorecards.

Positioning

Inference margin inversion — Same-day list cuts on mid/high tiers with company claims that serving compute fell enough to fund them; theme still breaks only on disclosed lab gross-margin compression across cuts, which neither post provides.

Connects to

Accenture embedded evaluator — Opus 5.5 cites METR/Frontier Design pre-release testing; governance design still the contested piece (news).

Baker pacing / compute margins — Price/efficiency drops can reallocate spend rather than destroy it (discussion).