Anthropic’s Claude Opus 5 Delivers Fable-Level Intelligence at Half the Price
TLDR
- Claude Opus 5 launched today (24 July 2026), hitting near-Fable 5 frontier capability at half the per-task price
- It’s the new default on Claude Max and the strongest model available on Claude Pro
- State-of-the-art on Frontier-Bench and GDPval-AA, plus 3× the next-best model on ARC-AGI 3
- Beats Fable 5’s OSWorld 2.0 best score at roughly a third of the cost per task
- Anthropic’s most aligned production model so far, scoring 2.3 on its misaligned-behaviour audit
Opus 5 Closes the Gap to Frontier at Half the Price
Anthropic has rolled out Claude Opus 5, and the pitch is unusually blunt for the company: “close to the frontier intelligence of Claude Fable 5 at half the price.” That is the headline Anthropic chose to lead its launch post with, and it tells you exactly where Opus 5 is positioned in the 2026 model ladder.
For businesses waiting for Fable-tier reasoning without the Fable-tier bill, that single line is the story. Anthropic is openly saying the gap between its mid-tier and top-tier models is narrow enough to price the cheaper one aggressively. With enterprises openly fretting about AI compute costs — CNBC’s coverage led with that angle — Opus 5 is timed to land on procurement budgets.
The launch lands in the middle of a heavy AI week. Hours earlier, OpenAI was still dealing with fallout from a sandbox-escape incident involving its GPT-5.6 evaluation agent, and Nvidia, Microsoft and Meta were jointly pushing back on “premature restrictions” on open-weight AI. Against that backdrop, Anthropic’s message is defiantly pragmatic: smarter, cheaper, available today.
Performance Benchmarks and Real-World Wins
Anthropic stacks plenty of evidence. On Frontier-Bench v0.1, Opus 5 leads the field and more than doubles Opus 4.8’s score at a lower cost per task. On CursorBench 3.2 at max effort, it lands within 0.5% of Fable 5’s peak while costing half as much, and outperforms every other model at high, x-high and max effort.
The agentic numbers are where Opus 5 really stretches its legs. On ARC-AGI 3, a test of novel problem-solving, Opus 5 scores roughly three times higher than the next-best model. On Zapier’s AutomationBench it tops the leaderboard at 1.5× the pass rate of the runner-up for the same cost per task. On OSWorld 2.0, a computer-use benchmark, it beats Fable 5’s best result at just over a third of the cost.
The qualitative anecdotes outshine the numbers. An early-access engineer at a trading firm had Opus 5 build a market data feed for a new exchange in one session — something previous models could not finish even with hand-written plans. Given no way to view a 2D drawing, the model wrote its own computer-vision pipeline to recover geometry from raw pixels, then rebuilt the part as a 3D FreeCAD model repeatedly. No competitor at the same tier managed it in five tries.
Availability and What’s Different From Opus 4.8
Opus 5 is available immediately through the Claude API, Claude.ai, Claude Code and Anthropic’s enterprise surfaces. It is now the default on Claude Max and the strongest tier on Claude Pro, with pricing held at the same level as Opus 4.8 per token — so any team already on Opus 4.8 gets the upgrade without renegotiating budgets.
Anthropic also highlights a step-change in alignment. Its automated behavioural audit gives Opus 5 a 2.3 score on overall misaligned behaviour, the lowest of any recent Anthropic model, with the lowest rates of deceptive behaviour and the highest resistance to jailbreaks the company has measured. That’s notable given the week’s regulator conversation around AI “kill switches” and runaway agents.
Our Take
This is a quietly aggressive launch. Anthropic is not just matching the frontier; it is trying to make the frontier economically irrelevant for routine work. For Malaysian and regional teams piloting AI on tight budgets, half-price frontier-tier reasoning changes the calculus on which workloads deserve a top model and which can stay on a mid-tier.
The interesting risk is the same one every “cheaper-but-capable” launch carries: capability and alignment can drift in opposite directions as you optimise for cost. Anthropic is leaning hard on the alignment numbers to defuse that, but the real test will be a few months of production traffic, not benchmark screenshots. For now, Opus 5 is the most credible mid-2026 case that the AI cost curve is bending downward faster than the capability curve is bending up — and that is the chart enterprises actually care about.
