Opus 5 in One Sentence: Frontier‑Grade Work at Half the Price
Claude Opus 5 is Anthropic’s new general‑purpose AI model that delivers coding and knowledge‑work performance close to its frontier system Fable 5 while costing around half as much per token, deliberately targeting everyday development and enterprise workflows where AI model pricing and productivity now matter more than headline benchmark scores. Anthropic has released Claude Opus 5 as its latest artificial intelligence model, openly pitching it as a near‑Fable model that is far cheaper to run. Opus 5 is priced at USD 5 (approx. RM23.0) per million input tokens and USD 25 (approx. RM115.0) per million output tokens, identical to Opus 4.8 and roughly half the cost of Fable 5. It is already live across Claude.ai, the Claude API, Claude Code, and Claude Cowork, and is now the default on Claude Max and the strongest option on Claude Pro. The strategic message is clear: the frontier is no longer reserved for premium‑priced models.
Benchmarks Show a New Cost‑to‑Capability Curve for Coding and Knowledge Work
Anthropic is betting that Claude Opus 5’s coding performance comparison against Fable 5 and rivals will be decided on performance per dollar, not bragging rights alone. On Frontier‑Bench v0.1, Opus 5 more than doubles Opus 4.8’s score while costing less per task, signaling a sharp jump in output for the same budget. On CursorBench 3.2, at maximum effort, it lands within half a percentage point of Fable 5’s best result yet at roughly half the cost per task, and Anthropic says it beats every other model at a given cost across high, xhigh, and max effort settings. The standout metric is ARC‑AGI‑3: Opus 5 scores three times higher than the next‑best model on genuinely novel problems, while Opus 4.8 barely registers and GPT‑5.6 Sol trails. Anthropic claims Opus 5 outperforms Fable 5 on 10 of 14 benchmark tests and surpasses GPT‑5.6 Sol in selected evaluations for software engineering and knowledge work, letting users tune reasoning effort to balance intelligence, speed, and token use. In practical terms, enterprises now have a frontier‑tier coder and analyst whose token bill looks much more like a mid‑range model.
Where Opus 5 Bends the Frontier, and Where It Still Falls Short
Opus 5 is not a pure upgrade over every frontier AI model; its cost‑to‑capability ratio shines in mainstream coding and knowledge work, but trade‑offs are real. On agentic search, knowledge work, and novel problem solving, the new model comes close to or even exceeds Fable 5’s performance for roughly half the token cost. It also improves on Opus 4.8 across life sciences evaluations, with a double‑digit gain on organic chemistry tasks such as reading spectroscopy data. However, the model falls short of Fable 5 in specialized categories like answering legal questions and performing multidisciplinary reasoning without tools. It is also “substantially behind” Mythos 5 in exploiting cybersecurity vulnerabilities, even though it closes much of the gap on vulnerability discovery. This is by design: Anthropic stresses that Opus 5 is not meant to lead in high‑risk cybersecurity applications. Anthropic’s behavioral audit shows Opus 5 with the lowest misaligned‑behavior score of recent Claude models, displaying less deceptive behavior and avoiding more reckless actions than its predecessors. The frontier, in other words, is being re‑drawn around safe productivity rather than maximum offensive capability.
Pricing Pressure and the Battle for Enterprise AI Cost Efficiency
The release of Claude Opus 5 lands in a market that is suddenly obsessed with enterprise AI cost and performance per task. OpenAI’s GPT‑5.6 Sol and Moonshot’s Kimi K3 have both been marketed heavily on cost efficiency, while lower‑cost Chinese models such as Kimi K3 and GLM 5.2 have pushed US labs to treat pricing as a first‑class competitive lever. Anthropic’s answer is nuanced: Opus 5 does not undercut every rival on raw price—it remains priced above Sol and well above K3—but the company argues its performance‑per‑dollar curve wins once you consider how many tokens and turns each model needs to finish a task. According to one source, “Opus 5 more than doubles Opus 4.8’s score on Frontier‑Bench v0.1 while costing less per task,” a clean statement of this performance‑per‑dollar thesis. For developers, startups, and enterprises keen to keep cloud bills in check, Claude Opus 5’s pricing of USD 5 (approx. RM23.0) per million input tokens and USD 25 (approx. RM115.0) per million output tokens could make advanced models more accessible without stepping down to budget‑tier AI. The net effect is clear: the competitive dynamic is shifting from “who has the biggest frontier model” to “who delivers the most useful work per dollar spent.”
What This Means for Developers, Teams, and the Next Phase of Claude
For developers and teams evaluating AI tooling, Opus 5’s arrival forces a more disciplined conversation about value: not just model quality, but how much usable work you can buy per unit of spend. Anthropic has positioned Opus 5 as a high‑performance AI model for everyday coding and knowledge tasks, aiming squarely at price‑sensitive segments that have been leaning on more expensive frontier AI models. This launch also follows a turbulent period in which access to Fable 5 was temporarily blocked worldwide after government concerns over its offensive cybersecurity capabilities, then restored with restrictions; the safer profile of Opus 5 is a direct response. On the product side, two platform updates ship in beta with Opus 5: developers can now swap tools mid‑conversation without breaking the prompt cache, and automatic fallbacks let flagged requests on Opus 5 or Fable 5 route to another model instead of being blocked outright. Put together, Opus 5 and these features give teams a path to frontier‑grade coding and knowledge work that is cheaper, safer, and operationally smoother. The frontier era is not ending—it is being priced into everyday development and office work, and Anthropic clearly wants Opus 5 to be the default choice when finance teams start asking hard questions about AI spend.
