Anthropic Releases Claude Opus 5 with 1M-Token Context and $5/$25 Per-Million-Token Pricing

Anthropic has released Claude Opus 5, positioned as the company's flagship model for reasoning, coding, and long-horizon agentic work. The model is listed on OpenRouter with a release date of July 24, 2026, a 1 million token context window, and pricing of $5 per million input tokens and $25 per million output tokens.

What's New

According to Anthropic, Claude Opus 5 is tuned for end-to-end software engineering tasks, including code review and bug detection, alongside visual analysis of charts and documents and generation of complex office deliverables. The company also claims the model can coordinate parallel subagents — multiple instances of the model working on different parts of a task simultaneously — a capability aimed at enterprise and developer workflows that require orchestrating multi-step, multi-agent pipelines.

Anthropic states the model maintains strong instruction-following and tool-use performance across extended task horizons. It also supports a lower-effort operating mode intended for workloads that prioritize latency and token efficiency over maximum reasoning depth, suggesting Opus 5 ships with adjustable compute/reasoning settings similar to prior Claude releases.

Context Window and Pricing

The model supports a 1 million token context window, matching the upper end of what leading frontier models currently offer. List pricing is $5 per million input tokens and $25 per million output tokens. OpenRouter notes that effective pricing — the average amount customers actually pay after prompt caching — can run 60–80% cheaper than list price depending on how much repeated context is sent in a workload, based on rolling 30-day data from the routing platform.

No official benchmark scores were included in the source listing. Anthropic has not published specific benchmark figures (e.g., SWE-bench, MMLU, or agentic coding evaluations) alongside this release, so performance claims beyond the qualitative description above remain unverified.

Availability

Claude Opus 5 is available through multiple hosting providers via OpenRouter, which routes requests based on selected mode: Balanced (price and speed), Nitro (fastest), or Exacto (highest tool-calling accuracy). The model is accessible through an OpenAI-compatible API, allowing existing SDK integrations to switch to it by changing the model slug.

What This Means

Opus 5's 1 million token context window and multi-agent coordination focus signal Anthropic is targeting enterprise agentic workflows and large-codebase engineering tasks rather than chasing raw benchmark leaderboard positions. The pricing — $5/$25 per million tokens — sits at a premium tier consistent with previous Opus-class models, but the emphasis on prompt caching discounts (60–80% off list price) suggests Anthropic and OpenRouter expect most production usage to involve heavily repeated context, such as large codebases or persistent agent sessions, where caching economics matter more than headline pricing.

The absence of published benchmark scores at launch is notable. Buyers evaluating Opus 5 against competitors like GPT-5-class models or Gemini's long-context offerings will need to run their own evaluations until Anthropic or third parties publish comparative data. The claimed subagent coordination capability, if it holds up under independent testing, would differentiate Opus 5 in the growing market for autonomous coding and office-automation agents — but until independent benchmarks arrive, that remains a company claim rather than a verified capability.