Anthropic Just Dropped Opus 4.5...
•November 24, 2025
0
Why It Matters
Opus 4.5 gives enterprises a higher‑performing, more token‑efficient coding assistant, but its premium pricing forces a new cost‑benefit calculus that could shift adoption toward models that balance raw capability with operational expense.
Summary
Claude Opus 4.5 arrived less than a week after Gemini 3 and Codex Max, positioning Anthropic at the top of the current frontier of coding‑focused large language models. The video walks through the model’s headline benchmark – a 80.9 % score on the SUI Bench coding test – which outpaces its predecessor Sonnet 4.5 (77.2 %) and beats rivals Gemini 3 Pro (76.2 %), GPT‑5.1 (76.3 %) and Codex Max (77.9 %). Across a suite of other tests, Opus 4.5 leads in TerminalBench (59.3 %) and T2 Bench tool‑use (98.2 % and 88.9 %), but falls short on GPQA‑Diamond, MMMU and multilingual Q&A where Gemini 3 retains the edge.
The presenter highlights several concrete data points: Opus 4.5’s pricing of $5 per million input tokens and $25 per million output tokens is roughly 50‑100 % higher than Gemini 3 Pro’s $2/$12 rates; yet the model consumes about half the context window to achieve comparable or better accuracy (12 k tokens vs 22 k for Sonnet 4.5). A striking anecdote is that Opus 4.5 outperformed every candidate on Anthropic’s notoriously difficult two‑hour take‑home engineering exam. In a T2 Bench airline‑service scenario, the model creatively upgraded a cabin before modifying a flight – a solution the benchmark marked incorrect, underscoring the model’s emergent reasoning beyond current test designs.
Anthropic also unveiled “advanced tool use” features to mitigate context‑window bloat from massive MCP tool libraries. A new tool‑search capability lets Claude retrieve only the necessary tool definition on demand, slashing context consumption from roughly 40 % to 5 % when loading dozens of integrations (e.g., GitHub, Slack, Grafana). Early adopters such as Dan Shipper and Ethan Malik praise Opus 4.5 as the best coding model they’ve used, noting dramatic gains in practical tasks like generating PowerPoints from Excel and one‑shot poetry tests.
The rollout signals a sharpening competitive race in AI‑assisted software development. Enterprises must weigh Opus 4.5’s superior coding and agent performance against its premium price, while the tool‑search innovation could lower operational costs for complex workflows. If Anthropic’s efficiency gains hold, developers may gravitate toward a model that delivers higher “intelligence per token,” potentially reshaping the economics of AI‑driven development platforms and pressuring Google and OpenAI to accelerate similar context‑optimisation features.
Original Description
Warp is free to try but for a limited time, try Warp Pro free for 7 days with 2,500 Al credits-no card required. Use my link ➞ https://go.warp.dev/bermanytagents
Download The Subtle Art of Not Being Replaced 👇🏼
Download Humanities Last Prompt Engineering Guide (free) 👇🏼
Join My Newsletter for Regular AI Updates 👇🏼
Discover The Best AI Tools👇🏼
My Links 🔗
👉🏻 Forward Future X: https://x.com/forward_future_
👉🏻 Instagram: https://www.instagram.com/matthewberman_ai
👉🏻 Discord: https://discord.gg/xxysSXBxFW
👉🏻 TikTok: https://www.tiktok.com/@matthewberman_ai
Media/Sponsorship Inquiries ✅
Links:
0
Comments
Want to join the conversation?
Loading comments...