Anthropic's Claude Opus 5.5 Matches Fable 5.1 at 40% Lower Cost

Anthropic released Claude Opus 5.5 on Tuesday, pricing the model at $4 per million input tokens and $20 per million output tokens, a 20 percent cut from Claude Opus 5's base rate on the same measure. Cached reads fall further, to 20 cents per million tokens, a 60 percent cut.

Anthropic said the new model performs at roughly the level of Claude Fable 5.1, the company's largest model, released in August, while running about 40 percent cheaper on a typical workload and generating output more than 30 percent faster than Opus 5. The claim rests on Anthropic's own benchmark runs, published alongside the model.

What Anthropic's own numbers show

On Terminal-Bench 4.0, a test of multistep command-line tasks, Anthropic scored Opus 5.5 at 66.4 percent against 55.8 percent for Fable 5.1, according to figures the company published and that SiliconANGLE reported independently the same day. On GDPval-AA v2.1, a knowledge-work evaluation scored in Elo terms, Opus 5.5 reached 1,846 against Fable 5.1's 1,735.

1. Claude Opus 5.566.4 %
2. Claude Fable 5.155.8 %

Source: Anthropic, Introducing Claude Opus 5.5, accessed 2026-09-23

Mario Rodriguez, GitHub's chief product officer, said in Anthropic's release materials that across Copilot CLI and VS Code, "Claude Opus 5.5 used among the fewest tokens and steps we measured," and that in VS Code it "solved more terminal tasks than Opus 5 in less than half the steps." GitHub's own changelog entry made the model available the same day to Copilot Pro+, Business and Enterprise users across nine surfaces, from Visual Studio Code to Xcode.

A price war with one day's distance

Opus 5.5 shipped hours before OpenAI released GPT-6 Sol and GPT-6 Luna, cutting its own API prices in half: Sol to $2 per million input tokens and $10 per million output, Luna to 10 cents and 50 cents. On a straight base-rate comparison, Sol is still cheaper than Opus 5.5 even after Anthropic's cut. Neither company has published a controlled benchmark pitting the two head to head.

Model	Input, per 1M tokens	Output, per 1M tokens
GPT-6 Luna	$0.10	$0.50
GPT-6 Sol	$2	$10
Claude Opus 5.5	$4	$20
Claude Opus 5	$5	$25
Claude Fable 5.1	$10	$50

The two releases extend a run of AI-beat coverage on Claude Fable 5.1's own launch terms and on Cognition's price-versus-benchmark tradeoffs against that same model, a comparison Anthropic's Tuesday numbers now reopen with a cheaper model of its own.

What Anthropic did not publish

The release page gives no parameter count, no context-window figure and no description of training data, consistent with Anthropic's practice for prior Claude releases. Anthropic also did not say when the cheaper Sonnet 5.5 and Haiku 5.5 models it has promised will ship, only that they are due after Opus 5.5.

Safety changes shipped with the model

Anthropic said Opus 5.5 was about 85 percent less likely than Opus 5 or Mythos 5.1 to attempt to bypass containment boundaries in a dedicated evaluation, and that the model keeps "preserved thinking," the anti-distillation safeguard Anthropic first shipped with Fable 5.1. External testers METR and Frontier Design evaluated the model before release, according to Anthropic; neither organization has published its own findings separately as of this writing.

Opus 5.5 is available now through the Claude API, on Amazon Web Services, Google Cloud and Microsoft Azure, and across Anthropic's own applications. Sonnet 5.5 and Haiku 5.5 are next, Anthropic said, without a release date.