Claude Opus 4.5 Released: Tops Coding Benchmarks While Slashing Prices

Anthropic released Claude Opus 4.5 on November 24, 2025, completing the 4.5 family (Sonnet 4.5 shipped in September, Haiku 4.5 in October). Per CNBC, the model excels at coding, computer use and complex enterprise tasks, topping SWE-bench Verified.

The surprise was pricing: Opus 4.5 API costs dropped to $5/$25 per million tokens — down more than two-thirds from the prior Opus — signaling that top-tier models no longer command top-tier prices, aimed squarely at enterprise-scale deployment.

It capped 2025's endgame: Gemini 3, Opus 4.5 and GPT-5.1 landed within two weeks of each other — frontier generations now turn over in weeks.

The Competitive Logic Behind the Price Cut

Dropping the flagship from the prior Opus's $15/$75 to $5/$25 is not generosity — it is positioning. In those two November weeks, Gemini 3 and GPT-5.1 arrived back to back (see our GPT-5.1 coverage), putting all three labs in direct contact on coding and agent workloads. Anthropic's revenue leans heavily on API and enterprise customers, and coding is its home turf — model bills from tools like Cursor and GitHub Copilot are acutely price-sensitive (see our Cursor 2.0 coverage). Pushing the flagship into mid-tier pricing forces rivals to choose between benchmark leadership and margin defense.

The immediate effect for developers: the old tiered habit — Sonnet for daily work, Opus only for hard problems — starts to dissolve, as a flagship model enters always-on cost territory for the first time.

A Three-Month, Three-Launch Cadence

From Sonnet 4.5 in September (see our coverage) to Haiku 4.5 in October to Opus 4.5 in November, Anthropic refreshed its entire family in a single quarter — far faster than the old annual-release rhythm. The intent is plain: with the AI coding tools war at full heat (see our coverage), a steady drumbeat of benchmark wins keeps the 'Claude for code' mindshare locked in, denying competitors any narrative window.

Our Take

Opus 4.5 matters more as a signal than as a score: frontier competition has shifted from 'who is stronger' to 'who is stronger and cheaper.' Once two-thirds price cuts become routine weaponry, model-layer margins keep compressing and value migrates up the stack to tools and agent orchestration — consistent with our call in the AI coding tools annual review. For users it is pure upside; for startups, 'cheaper than the flagship' is no longer a defensible position.