On September 22, Anthropic released Claude Opus 5.5 and OpenAI shipped GPT-6 Sol and GPT-6 Luna roughly 90 minutes later, collapsing the token-price floor for frontier-tier agentic work in a single afternoon. The pricing curves each lab had been defending for most of 2026 didn't drift downward. They cracked.

Opus 5.5 lists at $4 per million input tokens and $20 per million output tokens, a 20% cut from Opus 5's headline rates but roughly 40% cheaper on typical workloads once Anthropic's per-task token efficiency is folded in. Cache-write pricing drops from $6.25 to $5 per million tokens; cache-read from $0.50 to $0.20. The predecessor's $25 output price is gone.

GPT-6 Sol undercuts it directly, at $2 input and $10 output per million tokens, exactly half of Opus 5.5's uncached rates and 50% below GPT-5.6 Sol in both directions. GPT-6 Luna sits further down the ladder at $0.10 input and $0.50 output, a 58.3% output-side cut against its predecessor. An OpenAI spokesperson confirmed to VentureBeat that both rates are permanent rather than promotional.

The benchmark claims, predictably, don't converge. Anthropic reports Opus 5.5 scoring 40.0% on an early-access Zapier evaluation, and asserts it runs a software development benchmark at roughly one-third the cost of GPT-5.6 Sol, a comparison already stale by the time it published. OpenAI counters with 33.2% for GPT-6 Sol on AutomationBench 1.0.6 at xhigh effort, at $0.27 per task. Different harnesses, different tasks, no shared referee. As always, the same-harness comparison arrives weeks after the pricing does.

For a five-to-thirty-person team running automated outreach, code review, or content pipelines on either API, the arithmetic is unambiguous. Any workflow still billed at September 21 rates is overpaying by 40–60% at the token layer. The choice isn't whether to migrate but which lab's price-quality curve to migrate onto, and that answer depends on evals the labs haven't run against each other yet. Our same-afternoon breakdown walks through the head-to-head math.

Two second-order signals matter more than the sticker prices. Anthropic began rejecting the older computer_20251124 tool on the Claude API, tightening the cybersecurity posture around agentic browsing at the same moment it opened the price floor. And Claude Sonnet 5.5 and Haiku 5.5 are already telegraphed as forthcoming, which means Opus 5.5's pricing is the ceiling of what's coming, not the floor.

Price competition at the frontier used to be a quarterly event. It's now a Tuesday.

Sources

https://techcrunch.com/2026/09/22/anthropic-releases-opus-5-5-with-lower-prices-and-fable-level-performance/
https://venturebeat.com/technology/anthropic-releases-claude-opus-5-5-beating-fable-5-1-on-key-agentic-benchmarks-at-60-cheaper-api-price
https://venturebeat.com/technology/openai-releases-gpt-6-sol-and-luna-models-slashing-api-costs-50-or-more
https://www.theverge.com/ai-artificial-intelligence/998868/anthropic-claude-opus-5-5-cybersecurity
https://www.bnnbloomberg.ca/business/artificial-intelligence/2026/09/22/anthropic-unveils-claude-opus-55/