Loading the catalogue…
Loading the catalogue…
Anthropic's frontier Claude Opus 4.5: a proprietary, text-and-image model built for complex coding, research, and long-document reasoning with optional extended thinking.
Overview. Claude Opus 4.5 is Anthropic's flagship proprietary model, released November 2025, built for complex coding, research, and long-document reasoning. It accepts text and image input and returns text, with a 200K-token context window and an optional extended thinking mode for deeper reasoning on demand. Developers reach for it when a task genuinely needs frontier-tier reasoning depth rather than a fast, lightweight response.
Key strengths. Opus 4.5 scores 0.87 on GPQA-diamond and 0.74 on MMMU-Pro, reflecting strong performance on graduate-level science reasoning and multimodal understanding. Its terminalbench_hard score of 0.47 demonstrates competence in sustained, multi-step agentic tool-use inside a terminal environment. Anthropic positions it as the deepest capability tier in the Claude Opus 4 family, and its Arena standing has historically placed it well above typical frontier cutoffs.
When to use it. Use Opus 4.5 for complex coding tasks, research synthesis, and reasoning over long documents where the 200K context window matters. Turn on extended thinking when a problem needs deliberate multi-step reasoning rather than a quick answer; leave it off for everyday interactive use through the API, Claude apps, or supported cloud platforms. It's a strong fit for agentic tool-calling workflows that require the model to track state across several steps.
Limitations. Context is capped at 200K tokens with a 64K-token maximum output, and the model's broader training data extends only to around May 2025. Extended thinking improves reasoning quality but shifts response behaviour away from fast, interactive latency toward slower, more deliberate output. The model is text-and-image in, text-only out — it does not accept or produce audio.
Sovereignty note. Anthropic is a US-based creator, and Inference on Opus 4.5 carries CLOUD Act exposure as a result. Anthropic's open-weights policy is mixed, and this model itself is closed-weight and proprietary — organisations with strict data-residency requirements should factor this into their routing decisions via the Smart Router.
| Inference Provider | Sovereignty | Runs in | In / out per 1M | Status | Pin |
|---|---|---|---|---|---|
| L0 | €4.54 / 22.69+2 rates | Online |
12 use cases classified by the curator. The note says why.
Tool-calling, extended-thinking, and a 200K context window make this a strong agentic coding choice, though it lacks the 1M-token window that gives Opus 4.6/4.7 an edge on the largest repositories.
200K context accommodates most in-context document sets, though it falls short of the 1M window on Opus 4.6/4.7; extended-thinking supports grounded, careful answer generation.
Extended-thinking and a 200K context window support cross-document audit reasoning, though this variant sits below Opus 4.6/4.7's 1M-token ceiling for the very largest audit trails.
Extended-thinking supports deep, nuanced code review including security and architectural feedback; 200K context covers most codebases, though it lacks the 1M-token window of Opus 4.6/4.7 for whole-monorepo review.
200K context handles substantial reference sets, though very large corpora benefit more from the 1M-token Opus 4.6/4.7 variants; extended-thinking still supports structured comparative analysis.
Extended-thinking plus a 200K context window (not the 1M window of later Opus 4.6/4.7 releases) still supports deep multi-step decomposition; Claude Opus 4.5 remains a top-tier reasoning variant within the family.
Sourced numbers, not our own evaluations. Percentiles are against the whole Stav catalogue.
Computed from aa-intelligence-index, gpqa-diamond, aa-engineering-index, aa-coding-index, livecodebench, terminal-bench-v4, ifbench, aa-multilingual-index, aa-lcr.
curl https://api.stav.ai/v1/chat/completions \
-H "Authorization: Bearer $STAV_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "anthropic/claude-opus-4-5-20251101",
"reasoning_effort": "high",
"messages": [{"role": "user", "content": "Hello!"}]
}'→ X-Stav-Request-Id · X-Stav-Model · X-Stav-Provider → X-Stav-Sovereignty-Level — the rung the request was served at