Anthropic Debuts Claude Opus 5 With Cheaper AI Coding Power

What happened

Anthropic has released Claude Opus 5, positioning the model as a leaner, cheaper alternative that still delivers strong performance on coding and reasoning-heavy work. Rather than chasing raw peak performance, the company is emphasizing efficiency: how much intelligence customers get per dollar and per token spent 1. That framing marks a shift in how Anthropic is marketing its frontier models, moving away from pure capability benchmarks toward cost-adjusted value.

A central feature of the release is a new adjustable "effort" setting, which lets developers dial reasoning intensity up or down depending on whether they need maximum accuracy or faster, cheaper responses 1. This gives customers direct control over the tradeoff between speed, cost, and quality — a tradeoff that previously was baked into model selection rather than exposed as a user-facing dial.

Early enterprise customers have backed up Anthropic's efficiency pitch with concrete numbers. Harvey, a legal AI company, reported that Opus 5 matched the performance of Opus 4.8's maximum-reasoning mode while producing 26% fewer tokens on average, according to Niko Grupen, the company's head of applied research 1. Separately, Richard Pham of Fundamental Research Lab said that on difficult financial-modeling tasks, Opus 5 delivered roughly nine percentage points higher accuracy than its predecessor while needing about a third fewer turns and tool calls, and cutting completion time by 60% 1.

Reuters, reporting on the launch, framed Opus 5 differently — as a model that approaches the capability of a more powerful sibling model, which it identifies as "Fable 5," while costing half as much 2. That account places Opus 5 within a broader lineup strategy, positioning it as the value-oriented option beneath a flagship model, rather than emphasizing the token-level efficiency gains highlighted elsewhere.

Why it matters

The release lands amid what Reuters describes as an accelerating AI funding frenzy, underscoring how competitive pressure in the industry is pushing companies to differentiate not just on raw capability but on cost efficiency 2. As AI labs pour enormous sums into training ever-larger models, being able to offer comparable performance at a fraction of the price becomes a meaningful competitive lever, particularly for enterprise customers running high-volume workloads like legal document review or financial modeling.

The adjustable effort setting also reflects a broader industry trend: giving businesses fine-grained control over inference costs rather than forcing an all-or-nothing choice between a cheap, weaker model and an expensive, powerful one. For companies like Harvey and Fundamental Research Lab, that flexibility appears to translate directly into faster turnaround and lower operating costs on tasks that would otherwise require the most expensive, highest-effort model settings.

Where the reporting agrees

Both accounts agree on the fundamental story: Anthropic has launched Opus 5 as a lower-cost model that delivers performance close to its more capable counterpart 12. Both frame the release around cost efficiency rather than raw capability gains, suggesting Anthropic itself is steering coverage toward that narrative. The consistency between a detailed trade-press account and a wire-service report suggests the cost-efficiency positioning is coming directly from Anthropic's own messaging around the launch.

Where it doesn't

The two accounts diverge notably on what Opus 5 is being compared against. VentureBeat's account benchmarks Opus 5 against Anthropic's own prior model, Opus 4.8, citing customer-reported gains in token efficiency and accuracy on specific tasks 1. Reuters, by contrast, describes Opus 5 as nearing the capability of a separate, more powerful model it calls "Fable 5," at half the price 2 — a name and comparison that does not appear anywhere in the other account. This is a significant discrepancy: either Reuters is referring to an internal or codenamed Anthropic model not mentioned in the VentureBeat coverage, or there is a naming inconsistency in one of the reports.

Given that VentureBeat's reporting includes specific, named enterprise sources with granular performance figures tied explicitly to Opus 4.8, that account carries more evidentiary weight on the direct comparison points. The Reuters reference to "Fable 5" reads as either a distinct product line or a possible transcription issue, and the current reporting does not resolve which. Readers should treat the Opus 4.8 comparison as the better-supported baseline, while the "Fable 5" claim remains an open question pending further confirmation.

The bigger picture

Taken together, the coverage points to Anthropic doubling down on cost-per-performance as its competitive pitch, backed by real customer testimony rather than just internal benchmarks. Whether or not the "Fable 5" reference points to an actual forthcoming model, the throughline across both reports is unmistakable: Anthropic wants Opus 5 known as the model that gets enterprises most of the way to top-tier performance without top-tier pricing.