Claude Opus 5: The First Frontier Model Built Around Cost-Performance
Claude Opus 5 is Anthropic’s latest frontier-scale AI model that combines state-of-the-art benchmark performance with aggressive, lower pricing, aiming to become the default choice for enterprise buyers who care more about cost per unit of intelligence than about raw scores alone. Claude Opus 5 has debuted at the top of the Artificial Analysis Intelligence Index with a score of 61, nudging past Claude Fable 5’s 60 and pushing GPT-5.6 Sol into third place at 59. It is available today as a standalone model through the API, rather than only through a fallback tier, which means teams can adopt it directly without complex routing setups. The headline is not that Anthropic has edged out its own flagship; the story is that it has done so while saying Opus 5 costs half as much per task as Fable 5. In a market where top models are tightly clustered on capability, that price shift is the real disruption.
Benchmarks Show A Clear Lead In Knowledge Work And Coding
If you look past the single-point lead on the Artificial Analysis Intelligence Index, Claude Opus 5’s edge becomes much clearer in the benchmarks that matter for enterprise AI costs and day-to-day workflows. On GDPval-AA v2, which measures agentic professional knowledge work, Opus 5 at max effort hits 1861 Elo — more than 100 points ahead of both Fable 5 and GPT-5.6 Sol and 114 points ahead of Fable specifically. On AA-Briefcase, a proprietary test of agentic knowledge work, it scores 1720 Elo, a 146-point lead over Fable 5. That is not a marginal win; it is a wide gap in the kind of work enterprises pay for. Coding tells the same story. Inside Claude Code at extra-high effort, Opus 5 takes joint first place on the Coding Agent Index and posts the best score yet on SWE-Atlas-QnA, while matching GPT-5.6 Sol on Terminal-Bench v2.1 at 89% when both are pushed to high effort.
Claude Opus 5 Pricing Turns Intelligence Into A Budget Line Item
Anthropic is not shy about where it wants to compete: on Claude Opus 5 pricing and enterprise AI costs, not just bragging rights. Opus 5 keeps the same explicit token rates as Opus 4.8 — USD 5 (approx. RM23) per million input tokens and USD 25 (approx. RM115) per million output tokens — while promising that a typical task costs half of what it would on Fable 5. Cache writes sit at USD 6.25 (approx. RM29) per million tokens with a brief time to live, and cache hits drop to USD 0.50 (approx. RM2.30) per million. Combined with effort controls ranging from low to max, and a 1 million token context window, buyers can dial Opus 5 up or down across a 407 Elo span on GDPval-AA v2, with output token usage scaling roughly 8x from low to max effort. According to Artificial Analysis, Opus 5’s efficiency gains are most visible at the high end of the intelligence-versus-cost curve, which is exactly where serious enterprise deployments live.
Imminent Rollout Signals A New Default For Enterprise Buyers
The timing of Opus 5’s rise to the top of AI model benchmarks is not accidental; it comes as Anthropic’s Opus tier has been squeezed from below by Sonnet 5’s near-Opus performance at lower cost and from above by Fable and Mythos. Anthropic now looks close to shipping Claude Opus 5 widely, with signals pointing past internal testing into the hands of limited partners and early preparations spotted again by independent watchers. A brief listing in a coding tool and a matching model string on a major cloud provider’s quotas catalog suggest that rollout plumbing is already in place. Should it land fully, Opus 5 is expected to appear in the Claude apps for paid tiers, inside Claude Code, on the Claude Platform, and across Anthropic’s three cloud partners, most likely replacing Opus 4.8 instead of sitting beside it. For Max, Team, and Enterprise customers, that offers a clear path to higher Anthropic performance without stepping up to Mythos pricing.
A Cost-Per-Point Era That Puts Pressure On Every Lab
Claude Opus 5’s debut at the top of the Intelligence Index is notable, but the more important shift is philosophical: when six labs now have models above the 50-point threshold and the gap from first to fifth is within single digits, the game changes from raw score to cost per point of intelligence. Artificial Analysis highlights that Opus 5’s efficiency gains are most pronounced at the high end; at lower effort levels, its cost per task for a given intelligence level sits just behind the GPT-5.6 family, which keeps the competition honest. Opus 5 does show weaknesses: its factual knowledge still trails Fable 5 on AA-Omniscience, and its hallucination rate on that benchmark climbs to 50% as it answers more often instead of deferring. Yet for agentic knowledge work and coding — the workloads that drive enterprise AI bills — Anthropic now leads by a wide margin. In this compressed frontier, the model that delivers a score of 61 at lower cost than a 60-point rival is the one redefining the market, and for the moment, that model is Opus 5.
