Claude Opus 4.6 Is Now 67% Cheaper: What It Means for Your AI Budget

Anthropic just cut the flagship Opus price from $15 to $5 per 1M input tokens. For teams that shelved Opus because of cost, the calculus has changed significantly.

When Anthropic released Claude Opus 3 in early 2024, it was the most capable model available — and at $15 per million input tokens, it was priced accordingly. Most teams evaluated it, concluded it was too expensive for production use at scale, and deployed Claude Sonnet or Haiku instead. Opus became the model you used for one-off complex tasks, not the model you ran your workloads on.

Claude Opus 4.6 changes that calculus. At $5.00 per million input tokens and $25.00 per million output tokens, the new Opus is priced 67% lower than its predecessor on both dimensions. That is not a marginal adjustment — it is a structural repricing that moves Opus from "aspirational" to "evaluable" for a much wider range of production workloads.

The Full Anthropic Lineup, Repriced

To understand what the Opus 4.6 repricing means, it helps to see the entire Anthropic model stack side by side. The new lineup creates a much more graduated cost curve than the old one, where the gap between Sonnet and Opus was so large it was effectively a binary choice.

* Legacy models (Opus 3, Opus 4.5) remain available but are no longer the recommended path for new workloads. Pricing from tokencost.is as of March 25, 2026.

The old lineup had a $12 gap between Sonnet ($3.00) and Opus ($15.00) on input tokens — a 5x price jump for the capability upgrade. The new lineup has a $2 gap between Sonnet 4.6 ($3.00) and Opus 4.6 ($5.00) — a 1.67x jump. That is a fundamentally different trade-off. Teams that previously chose Sonnet purely because Opus was unaffordable now have a legitimate decision to make.

What Workloads Benefit Most

Not every workload benefits from the Opus repricing equally. The models where the capability gap between Sonnet and Opus is most pronounced — and where that gap translates directly into business value — are the ones worth re-evaluating first.

Complex document analysis is the clearest case. Legal contract review, financial report summarization, technical specification parsing — these are workloads where Opus's stronger reasoning and longer coherent output windows produce materially better results than Sonnet. At $15/1M, the cost premium was hard to justify. At $5/1M, the math changes substantially.

Consider a legal tech team running 200 contract review requests per day, with an average of 8,000 input tokens and 2,000 output tokens per request. Here is what the monthly cost looks like across the Anthropic lineup at current pricing:

* Workload: 200 requests/day × 30 days, 8,000 avg input tokens, 2,000 avg output tokens. Pricing from tokencost.is.

At the old Opus pricing, this workload cost $9,720 per month — nearly $116,000 per year. At Opus 4.6 pricing, the same workload costs $3,240 per month. That is a $6,480 monthly saving just from the price drop, with no change to the model, the prompts, or the output quality. For teams already running Opus 4.5, this saving is automatic once they migrate to 4.6.

Where Sonnet Still Wins

The Opus repricing does not mean every workload should migrate to Opus. For high-volume, lower-complexity tasks, Sonnet and Haiku remain the correct choices. The cost differential still matters at scale — Opus 4.6 is 1.67x more expensive than Sonnet 4.6 on input and output tokens, and that multiplier compounds quickly at high request volumes.

A customer support bot processing 2,000 requests per day illustrates the point. At this volume, the cost difference between Sonnet and Opus is substantial even at the new Opus pricing:

* Workload: 2,000 requests/day × 30 days, 600 avg input tokens, 250 avg output tokens. Pricing from tokencost.is.

For a customer support bot, Claude Opus 4.6 costs $3,375 per month versus $576 for Haiku 3.5 — nearly 6x more expensive. Unless the quality difference produces measurable business value (lower escalation rates, higher CSAT scores that can be quantified), Haiku remains the correct choice for this workload type. The Opus repricing does not change that analysis.

The Decision Framework

The right way to think about the Opus 4.6 repricing is not "should we switch everything to Opus?" but rather "which workloads were previously Opus-quality requirements that we were running on Sonnet because of cost?" Those are the workloads to re-evaluate first.

What to Do Right Now

If your team is currently running any workload on Claude Opus 3 or Claude Opus 4.5, the migration path to Opus 4.6 is the highest-ROI action available to you today. The model is more capable, the API is compatible, and the price is 67% lower. There is no reason to remain on the legacy Opus pricing.

If your team is running complex workloads on Claude Sonnet because Opus was too expensive, now is the time to run a proper A/B evaluation. The quality gap between Sonnet and Opus is real and measurable on reasoning-heavy tasks. At the new Opus pricing, the cost premium may be justified by the output quality improvement — but you need real data from your specific workload to know for certain.

The worst outcome is making this decision based on intuition rather than numbers. The calculator below will show you the exact monthly cost of every model in the Anthropic lineup — and every other provider — for your specific workload parameters. Run the numbers before you commit.