Claude Opus 5: Near-Fable Power at Half the Cost

Opus 5 in One Sentence: Frontier-Like Intelligence Without Frontier Pricing

Claude Opus 5 is Anthropic’s newest large language model that aims to deliver performance close to its more expensive Claude Fable 5 in coding, knowledge work and agentic tasks while keeping the same low API pricing as Opus 4.8, making frontier-like capabilities available for everyday development and enterprise use. This is not a minor iteration; it is a deliberate strike at the idea that top-tier AI must come with top-tier invoices. By holding the line at USD 5 (approx. RM23) per million input tokens and USD 25 (approx. RM116) per million output tokens while outscoring Fable 5 on most reported benchmarks, Opus 5 redraws the map for LLM cost comparison and should make many CTOs question why they still pay premium rates for incremental gains.

Performance: When “Good Enough” Starts Looking Better Than “Best”

On paper, Opus 5 is supposed to be the affordable Anthropic Fable alternative; in practice, it often looks like the smarter economic choice even when Fable 5 is available. Anthropic reports that Opus 5 beats Fable 5 on 10 of 14 benchmarks and posts a 43.3% score on Frontier-Bench v0.1 compared with 33.7% for Fable 5 and 21.1% for Opus 4.8, with GPT-5.6 Sol at 34.4%. That means many real-world software and knowledge workflows can be handed to Opus 5 without feeling like you’ve downgraded. The quotable line here is simple: “On several coding and knowledge work evaluations, Opus 5 is the new state-of-the-art.” Yes, Fable 5 still wins on some deep coding and exam-style tests, and Mythos 5 remains ahead in offensive cybersecurity and long-running biology work. But for the bulk of everyday development and business automation, paying double for a few extra percentage points starts to look more like vanity than strategy.

Pricing, Effort Settings and the End of One-Size-Fits-All AI Bills

The most underappreciated change with Claude Opus 5 pricing is not just the headline numbers; it is the new level of cost control Anthropic hands to developers. Keeping the API at USD 5 (approx. RM23) per million input tokens and USD 25 (approx. RM116) per million output tokens while doubling down on performance sends a blunt signal: the era of spiralling LLM bills for routine workloads is over if you choose wisely. The adjustable “effort” setting—where customers trade more reasoning and tokens against speed and cost—means enterprises can reserve maximum brainpower for complex edge cases instead of burning budget on every autocomplete. According to Anthropic, Opus 5 came within 0.5% of Fable 5 on CursorBench 3.2 at maximum effort while costing around half as much per task. For cost-conscious teams, that is precisely the kind of quotable procurement math that wins budget approvals: near-frontier quality on frontier-style tasks, but priced like a mid-range tool.

Enterprise Reality: Cheaper, Less Locked-Down, and No Data Retention Hooks

Where Opus 5 most clearly challenges the status quo is in enterprise AI models. It is now the default Claude Max model and the strongest option in Claude Pro, with safety controls that Anthropic expects to intervene about 85% less often than on Fable 5. In simple terms: fewer frustrating refusals on legitimate work, without drifting into Mythos-style risky territory. For enterprises obsessed with compliance and privacy, one detail matters more than any benchmark chart: general access to Opus 5 carries no data-retention requirements. That removes a whole category of hidden cost, from governance overhead to legal reviews, and makes Opus 5 easier to drop into sensitive workflows. Combined with its gains in business automation—such as a 26% AutomationBench pass rate versus 17.4% for Fable 5 and 17% for Opus 4.8—Opus 5 looks less like a budget model and more like a pragmatic default for serious, regulated environments.

Competitive Landscape: When Price Pressure Meets Frontier Ambition

The timing of Opus 5 is not accidental. US-based vendors are feeling pressure from lower-cost Chinese models such as Kimi K3 and GLM 5.2, which have turned LLM cost comparison into a boardroom agenda rather than a niche concern. Anthropic’s bet is clear: businesses will prioritize affordable AI performance and overall value over chasing the cheapest token price. By offering a model that often outperforms both its own flagship Fable 5 and rivals like GPT-5.6 Sol at a given cost, Anthropic has effectively drawn a line under “high-cost barrier” excuses for delaying AI adoption. Lower Costs Could Appeal to Businesses, but it is the combination of pricing, controllable effort levels, lighter safety friction and no data retention requirements that makes Opus 5 hard to ignore for developers and enterprises. The conclusion is straightforward: if your AI strategy still assumes that best-in-class means premium-only, Opus 5 is a signal that the market has shifted, and your procurement playbook needs to catch up.