Anthropic is rolling out Claude Opus 4.5, a new flagship model positioned as its best option yet for coding, long‑running agents, and computer use, with broader availability and sharply reduced prices across its platform and partner clouds.
Release date, price, and availability
Claude Opus 4.5 is available now in Anthropic’s own apps, through the Claude API under the model name claude-opus-4-5-20251101, and on Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry.
| Offering | Where it runs | Pricing (per 1M tokens) | Status |
|---|---|---|---|
| Claude Opus 4.5 (Claude API) | Claude Developer Platform | $5 input / $25 output | Generally available |
| Claude Opus 4.5 (Microsoft Foundry) | Azure (East US2, Sweden Central) | $5 input / $25 output (serverless pay‑as‑you‑go) | Public preview |
| Claude Opus 4.5 (Claude apps) | Claude Pro, Max, Team, Enterprise | Included in plan; higher Opus usage limits | Live |
Anthropic has also removed Opus‑specific caps for users with Opus 4.5 access and increased overall limits for Max and Team Premium plans, bringing Opus usage roughly in line with previous Sonnet allocations in the consumer apps.
New capabilities in Claude Opus 4.5
Anthropic describes Opus 4.5 as its most capable model so far, with state‑of‑the‑art performance on real‑world software engineering benchmarks. On the SWE-bench Verified evaluation of software bug‑fixing, Opus 4.5 reaches 80.9%, ahead of Claude Sonnet 4.5 and earlier Opus versions on Anthropic’s internal leaderboard.
The model is designed to handle:
- Advanced coding – multi‑day software projects compressed into hours, with improvements across languages, planning, and architecture.
- Agentic workflows – long‑horizon, multi‑step tasks that require planning, tool use, and adaptive decision‑making.
- Enterprise documents – generation and editing of spreadsheets, slides, and long‑form documents with consistent structure and tone.
Opus 4.5 also improves vision and computer use. On OSWorld, a benchmark for real‑world computer tasks, it delivers Anthropic’s best scores to date, enabling more reliable browser and desktop automation.
Anthropic highlights a 200K‑token context window and “hybrid reasoning” that supports both fast answers and extended thinking. An effort parameter on the Claude API lets developers trade off latency and cost against deeper reasoning: at medium effort, Opus 4.5 matches Sonnet 4.5’s best SWE-bench Verified score while using 76% fewer output tokens; at high effort, it surpasses Sonnet 4.5 while still using fewer tokens.
Coding, agents, and computer use improvements
Opus 4.5 targets production software engineering and agents that can work across many tools. Anthropic reports:
- Leading performance in 7 of 8 programming languages on SWE-bench Multilingual.
- A 10.6‑point improvement over Sonnet 4.5 on the Aider Polyglot coding benchmark.
- Higher scores on agentic search (BrowseComp-Plus) and long‑horizon task benchmarks such as Vending-Bench.
Customers testing Opus 4.5 describe efficiency gains on long‑running coding tasks, with some reporting up to 65% fewer tokens to reach the same or better results and noticeable reductions in tool‑calling and build errors. Anthropic’s own internal engineering take‑home exam, used for performance engineering candidates, saw Opus 4.5 achieve a higher score within a two‑hour limit than any human candidate so far, when used with parallel test‑time compute inside Claude Code.
The model’s agent behavior is designed to be more persistent and creative within constraints. On τ2-bench, a suite of multi‑turn agent tasks, Opus 4.5 can find non‑obvious but policy‑compliant solutions, such as restructuring steps in an airline booking problem by upgrading a ticket class before changing flight dates.
Updates to Claude apps and developer tools
Alongside the model, Anthropic is updating the broader Claude ecosystem to take advantage of Opus 4.5.
- Claude apps: Long conversations now automatically compact older context so chats can continue without hitting hard limits. The Claude for Chrome extension, which lets the model work across browser tabs, is open to all Max users, and Claude for Excel is in beta for Max, Team, and Enterprise tiers.
- Claude Code: Plan Mode now asks clarifying questions, generates a
plan.mdfile users can edit, and then executes against that plan. Claude Code is also available in the desktop app, allowing multiple local and remote coding sessions in parallel. - Developer Platform: Effort control, context compaction helpers, and advanced tool‑use features are now exposed to developers, enabling longer‑running agents with fewer interventions. Anthropic reports a roughly 15‑point gain on a deep research evaluation when combining these techniques with Opus 4.5.
On Azure, Microsoft Foundry wraps Opus 4.5 into its model catalog, GitHub Copilot paid plans, and Microsoft Copilot Studio, positioning it as an option for production agents that span coding, financial modeling, cybersecurity, and other workflows with heavy tool use.
Safety, alignment, and security
Anthropic classifies Opus 4.5 as its most robustly aligned frontier model so far, with lower “concerning behavior” scores on its internal alignment metrics than previous Claude releases. The Claude Opus 4.5 system card details evaluations across deception, sycophancy, power‑seeking, and other failure modes.
The model is also hardened against prompt‑injection attacks, where malicious instructions attempt to override intended behavior. Anthropic reports that Opus 4.5 is harder to trick than other frontier models on a strong external prompt‑injection benchmark developed by Gray Swan.
For enterprises weighing which Claude tier to use, Opus 4.5 now sits at the top of the lineup above Sonnet 4.5 and Haiku, aimed at the hardest coding and agent workloads while keeping cost and safety in line with large‑scale deployment.
As Anthropic and its cloud partners fold Opus 4.5 into more products, the model is likely to become the default engine behind many autonomous coding agents and office automations rather than a niche, premium option.
