Claude Opus 5.5 and GPT-6 Sol land less than two hours apart, and both labs cut prices

Anthropic says Opus 5.5 matches Fable 5.1 at $4 and $20 per million tokens, a fifth below Opus 5. OpenAI halved Sol and Luna to $2 and $10 and $0.10 and $0.50. An independent tester puts Opus 5.5 on top of its index and Sol level with its predecessor.

Agent-drafted, human-editedPublished September 23, 2026

Reviewed by Trevor Brown

Anthropic released Claude Opus 5.5 on September 22, the first model in a Claude 5.5 family, and cut its prices below Opus 5.1 OpenAI answered 101 minutes later with GPT-6 Sol and GPT-6 Luna at half the price of the GPT-5.6 models they replace.3 4 6

Why it matters: Both labs shipped the same day and both led with cost. Anthropic's Opus tier got cheaper, which it did not when Opus 5 launched in July, and OpenAI's mid and low tiers now undercut it by a wide margin. The two announcements compare against each other's older models, so neither one settles which is ahead.

What Anthropic claims: Opus 5.5 "performs at the level of Claude Fable 5.1 on most work" and, Anthropic says, produces output over 30 percent faster than Opus 5.1 Anthropic reports 66.4 percent on Terminal-Bench 4.0 against 55.8 for Fable 5.1 and 57.9 for GPT-6 Astra, a figure it takes from OpenAI, and 1,846 Elo on GDPval-AA against 1,735 for Fable 5.1.1 The company adds that "benchmark margins have become a less reliable guide to real-world differences."1

What it costs: $4 per million input tokens and $20 per million output, down from $5 and $25 for Opus 5. Cache reads fall to $0.20 per million, and Anthropic says the combination cuts typical workloads 40 percent.1 2 It is live on the API as claude-opus-5-5 and on AWS, Google Cloud and Microsoft Azure. Sonnet 5.5 and Haiku 5.5 "will follow in the coming weeks."1

The catch: Opus 5.5 carries cybersecurity safeguards similar to Fable 5.1's, so "most cybersecurity tasks will be re-routed to Opus 4.8."1 Anthropic ran its benchmarks with those safeguards on, and says cyber tasks that tripped them were completed by Opus 4.8, and biology and frontier-model development tasks by Opus 5.1 Anthropic calls it its first release since Dario Amodei's September 12 call to pace the frontier, and says METR and Frontier Design tested it before release.

Provenance & disclosure

AI use

Agent-drafted, human-edited

Disclosure

AI drafted this brief from Anthropic's and OpenAI's launch pages, both companies' pricing pages and X posts, and Artificial Analysis's published results and X post. Every number was checked against the page it came from. Capability and cost figures are each company's own claims unless attributed to Artificial Analysis. Trevor Brown reviewed and approved the story for publication.

Reviewed by

Trevor Brown — Editor-in-chief

Sources

1.Official sourceThe September 22 release as "the first model in our new Claude 5.5 family," the "our first release since we called for pacing the frontier" line and the "tested before release by external evaluators, including Frontier Design and METR" sentence, the link to the Opus 5.5 System Card, the "performs at the level of Claude Fable 5.1 on most work" and "more than 30% faster than Opus 5" claims, the $4 / $20 per million pricing "20% less than Opus 5," the $0.20 cache-read rate and the "cost 40% less than Opus 5 on typical workloads" claim, the Terminal-Bench 4.0 (66.4% against 55.8% for Fable 5.1 and 57.9% for GPT-6 Astra, the latter "as reported by OpenAI") and GDPval-AA v2.1 (1,846 Elo against 1,735) rows, the "benchmark margins have become a less reliable guide to real-world differences" sentence, the "most cybersecurity tasks will be re-routed to Opus 4.8" safeguard, the evaluation note that safeguarded cyber tasks "were completed by Claude Opus 4.8, and biology and frontier LLM development tasks were completed by Claude Opus 5," the `claude-opus-5-5` identifier, the AWS / Google Cloud / Microsoft Azure availability and the "will follow in the coming weeks" line for Sonnet 5.5 and Haiku 5.5, all read off Anthropic's launch page, September 23, 2026: Introducing Claude Opus 5.5.

The independent read: Artificial Analysis scored Opus 5.5 at 58 on its Intelligence Index at max effort, "the highest score we have measured by several points," while noting it used about 1.6 times the output tokens of Opus 5 to get there.7 On Sol and Luna it said scores "remain level with GPT-5.6, with progress in some evaluations and regressions in others."8

Context: Anthropic launched Opus 5 on July 24 at unchanged rates, then Fable 5.1 on September 1 at $10 and $50. OpenAI matched that price with GPT-6 Astra two days later. Tuesday's releases move the fight down-market.

What we still don't know: OpenAI's post sends readers to Astra's system card rather than a new one.4 Anthropic published an Opus 5.5 system card alongside the launch, which we have not yet read.1