Anthropic continues its push into the frontier-model market with the release of Claude Opus 4.5 — an updated flagship version of its AI that has become simultaneously smarter and significantly cheaper. Germany’s t3n notes that Opus 4.5 not only improved scores on key benchmarks but also became far more efficient in handling tokens, which directly challenges the positions of OpenAI and Google.
In technical evaluations, the model shone particularly brightly on SWE-Bench Verified — a standard for assessing programming in real-world conditions. Opus 4.5 reaches about 80.9%, surpassing OpenAI’s competitor (GPT-5.1-Codex-Max with roughly 77.9%), allowing Anthropic once again to claim the title of “the world’s best coding model.”
At the same time, the new version learned to think more economically: to achieve the same or even better results, it requires fewer tokens than its predecessor, and users can now themselves adjust the depth of reasoning via the Effort parameter. The API offers three effort levels: at medium, Opus 4.5 produces Sonnet-4.5-level quality while using up to 76% fewer output tokens; at maximum, it outperforms Sonnet by another ~4 percentage points while still remaining nearly twice as economical.
Anthropic is also striking hard on price. Access to Opus 4.5 via API now costs $5 per million input tokens and $25 per million output tokens — an aggressively low offer for a model positioned as “the world’s best AI for code, agents, and computer work,” even compared to GPT-5.1 and Gemini 3 Pro.
Anthropic’s strategy is becoming clear: Opus 4.5 is not about flashy experiments but about being a quiet yet extremely useful “workhorse” for business. Improved coding, Excel and presentation handling, management of complex agents, deep research, and a simultaneous reduction in compute cost — all of this turns the new model into a tool that major companies can realistically use every day. In the oversaturated frontier-model race, Anthropic is betting on a combination of intelligence and pragmatic economics — and that is precisely what hits competitors hardest right now.
