This Week in AI
  Claude Cookbook: Anthropic's Official Recipe Hub Goes Viral  Health in ChatGPT: OpenAI's Big Health AI Launch  Claude Opus 5: Anthropic's Step-Change for AI Agents  Claude Opus 5 Pricing: $1/MTok Replaces Opus 4  OpenAI's GPT-Red Explores AI Self-Improvement  Apple Sends Legal Letters to Dozens of OpenAI Staff  GLM 5.2, OpenHuman, Free Claude Code: New AI Tools  Thinking Machines Launches Inkling Open-Weights Model
LLM Launches & Updates

Claude Opus 5 Pricing: $1/MTok Replaces Opus 4

Anthropic's pricing page now lists Claude Opus 5 at $1/MTok, replacing Opus 4, with a 50% batch discount and a $20 per-seat plan. Here's what changed.

Claude Opus 5 Pricing: $1/MTok Replaces Opus 4

> **TL;DR:** Anthropic has updated its public pricing page: Claude Opus 5 is now listed at $1 per million tokens, taking over the exact price point previously held by Opus 4, whose listing has been removed. The page also adds a $0.50/MTok batch-processing rate — a 50% saving — and a $20 per-seat-per-month plan with usage billed on top.

Key Takeaways

- Claude Opus 5 is now listed at $1 per million tokens on Anthropic's pricing page — the same slot Opus 4 held before its listing was removed. - A new $0.50/MTok batch-processing rate offers a 50% discount for asynchronous, latency-tolerant workloads. - Anthropic now advertises a $20 per-seat-per-month plan with usage billed on top of the seat fee. - Budgets built around the old Opus 4 listing now buy the newer flagship model at the same per-token rate. - Confirm input/output direction and any tier conditions on the live pricing page before updating cost models.

Anthropic has updated its official [pricing page](https://www.anthropic.com/pricing), and the most consequential edit is a straight model swap: Claude Opus 5 now appears at $1 per million tokens (MTok), the exact price point previously occupied by Opus 4 — and the old Opus 4 listing has been removed entirely. Two further additions round out the refresh: a $0.50/MTok rate tied to a 50% batch-processing discount, and a $20 per-seat-per-month plan with usage billed on top of the subscription.

What Changed, Line by Line

Comparing the live page against our last stored snapshot (July 2026), the diff comes down to three additions and one removal:

- **Added:** Claude Opus 5 listed at $1/MTok, with the page positioning it for complex work - **Added:** a $0.50/MTok figure paired with a note about saving 50% through batch processing - **Added:** a plan at $20 per seat per month, with usage billed in addition to the seat fee - **Removed:** the Opus 4 entry at $1/MTok

That is the entire visible change — no other tiers moved in this snapshot. But each of the three additions tells its own story about where Anthropic is steering its commercial model.

Opus 5 Inherits the Flagship Price Point

The significant part of the swap is what didn't change: the number. By slotting Opus 5 into the $1/MTok position Opus 4 held, Anthropic is treating a generational upgrade as a like-for-like replacement rather than an occasion to reprice. For teams already budgeting around the old listing, the practical read is simple: the newer flagship now costs what the previous one did, and the previous one no longer appears on the public rate card. If your cost models or internal chargeback sheets reference Opus 4's listed price, they now effectively describe Opus 5.

One caveat worth stating plainly: a pricing-page snapshot captures headline figures, not the full rate card. Per-token pricing typically differs between input and output tokens, and can vary with context length, caching, and volume tiers. The $1/MTok figure is what the page displays for Opus 5; confirm the direction and any conditions on [Anthropic's pricing page](https://www.anthropic.com/pricing) before rewriting a budget.

Batch Processing at $0.50/MTok

The new $0.50/MTok line is presented as a 50% saving via batch processing — exactly half the $1/MTok headline rate. Batch APIs generally trade latency for cost: jobs are submitted asynchronously and results collected later, which suits evaluations, dataset labeling, bulk summarization, and overnight document pipelines rather than interactive chat. If a meaningful share of your workload can tolerate a delay, this discount functions as a blanket price cut for offline work.

$20 Per Seat, Plus Usage

The third addition is a $20 per-seat-per-month plan with usage billed on top. Hybrid seat-plus-usage pricing is fast becoming the default commercial shape for AI platforms: the seat fee covers access and collaboration, while metered billing keeps heavy users paying in proportion to what they consume. It also simplifies procurement — a predictable per-head line item with variable, API-style costs layered above it. Monetization experiments like this are spreading across the AI stack, from seat-based developer tooling to creator royalty programs such as the [ElevenLabs voice marketplace](https://speka.info/blog/elevenlabs-voice-marketplace-how-creators-earn-royalties).

![Conceptual illustration comparing standard and discounted batch token processing costs](https://supabase.srv1729373.hstgr.cloud/storage/v1/object/public/blog-images/speka-info/claude-opus-5-pricing-update-1-48d5d9927bc68dcc.png)

Why This Matters for Buyers

If you build on Claude, the short version is that the model behind the $1/MTok price just got better, and offline workloads gained a formal 50% discount path.

Per-token prices compound fastest in agentic workloads. Coding agents — from Claude Code itself to specialized tools like [Juggler, the GUI coding agent from the creator of JUCE](https://speka.info/blog/juggler-juce-creators-new-gui-coding-agent) — routinely consume millions of tokens per session as they read files, plan, and iterate. At that consumption profile, the gap between a flagship model at $1/MTok and the same model at $0.50/MTok in batch mode is the difference between an experiment and a budget line item.

The competitive backdrop matters too. Challenger models keep pressing on price — see our roundup of [GLM 5.2, OpenHuman, and free Claude Code offers](https://speka.info/blog/glm-5-2-openhuman-free-claude-code-new-ai-tools) — and holding the flagship tier at a stable per-token figure while swapping in a stronger model is how an incumbent answers that pressure without announcing a headline price cut.

![Abstract visual of seat-based subscription pricing combined with metered usage billing](https://supabase.srv1729373.hstgr.cloud/storage/v1/object/public/blog-images/speka-info/claude-opus-5-pricing-update-2-b44cee2fecc06218.png)

What to Watch Next

Pricing pages are living documents, and this snapshot leaves open questions. The diff doesn't say whether Opus 4 remains available at unlisted rates or is being phased out — the only verified fact is that it no longer appears at $1/MTok on the public page. Likewise, the full terms behind the $20 seat plan and the batch discount live on [Anthropic's pricing page](https://www.anthropic.com/pricing), which is worth reading in full before committing spend. We track these snapshots continuously; for the running log of model releases and price moves, follow our [LLM Launches & Updates](https://speka.info/llm-updates/) hub.

Frequently Asked Questions

How much does Claude Opus 5 cost?

Anthropic's pricing page currently lists Claude Opus 5 at $1 per million tokens, positioned for complex work. Check the page for input/output specifics and any tier conditions.

What happened to Claude Opus 4 on Anthropic's pricing page?

The Opus 4 listing at $1/MTok has been removed, and Opus 5 now occupies that price point. The snapshot doesn't indicate whether Opus 4 remains available at unlisted rates.

How does Anthropic's batch processing discount work?

The page advertises $0.50/MTok — a 50% saving versus the $1/MTok rate — via batch processing, which suits asynchronous, latency-tolerant workloads like evaluations and bulk summarization.

What is Anthropic's $20 per-seat plan?

It's a newly listed plan at $20 per seat per month, with usage billed on top of the seat fee. Full plan details are on Anthropic's pricing page.

Is the $1/MTok rate for input or output tokens?

The public snapshot shows the headline figure without specifying. Token direction, context-length tiers, and caching rates should be confirmed on Anthropic's pricing page.

Sources

- https://www.anthropic.com/pricing

← Back to all posts