Anthropic released Claude Opus 5 on 24 July 2026 at $5 per million input tokens and $25 per million output tokenshalf the price of its Fable 5 flagship, and paired the launch with a per-request effort toggle that turns model capability into a dial. Customer benchmarks arrived with the announcement: Zapier reports a 100% pass rate on account-health automationsBox measures an 8% lift over Opus 4.8and Lovable clocks a 22% gain over Opus 4.7.
Before this release, Anthropic's lineup forced a clean trade: pay frontier rates for Claude Fable 5 or take the capability discount of Opus 4.8 at $5/$25. Claude Opus 5 keeps the Opus 4.8 price point and, on Anthropic's own numbers, closes most of the distance to the flagship. The company calls it “a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price”by its own account the clearest step up in the Opus family since version 4.5. The timing lands mid-cycle in an inference market where price per unit of capability, more than raw capability, has become the axis of competition. For CTOs halfway through 2026 budget cycles, the release resets the reference price for frontier-adjacent capability at exactly the moment renewal negotiations open.
What changed: frontier-adjacent capability at the Opus 4.8 price
The core specs come straight from Anthropic's announcement. Standard pricing sits at $5 per million input tokens and $25 per million output tokensmatching Opus 4.8 and undercutting Fable 5 by half. A fast mode runs roughly 2.5 times faster at twice the base price. The new effort setting, low, medium or high, lets teams “optimize for intelligence or conserve tokens for faster and cheaper results” on a per-request basis, which means a single model identifier, claude-opus-5, now spans use cases that previously required two separate models. The model ships on Claude.ai, Claude Code, Claude Cowork, the Claude API and all cloud deployment options, and per TestingCatalog it takes over as the default model on Claude Max and the strongest model on Claude Pro.
The benchmark story is aggressive. Anthropic reports that Opus 5 leads Frontier-Bench v0.1 and more than doubles Opus 4.8's performance at lower cost. On CursorBench 3.2 at maximum effort it lands within 0.5% of Fable 5's peak score at half the cost. On ARC-AGI 3 it posts a score three times higher than the next-best modeland on OSWorld 2.0 it beats every rival at a given cost level and surpasses Fable 5 at one-third the cost. Scientific evaluations show gains of 10.2 percentage points in organic chemistry and 7.7 points on protein-sequence tasks versus Opus 4.8. These are vendor numbers and deserve independent verification, the pattern across them is consistent, though: near-flagship capability at mid-tier economics.
Customer data points give the claims commercial weight. Zapier CEO Wade Foster reports a 100% pass rate on end-to-end account-health automationsa workflow that defeated previous models. Box CTO Ben Kus measures an 8% overall lift versus Opus 4.8, with 11% on data analysis and 17% on due diligence. Lovable co-founder Fabian Hedin sees performance up 22% over Opus 4.7 with far less variance run to run. Devin CEO Scott Wu says the model “approaches Fable-level performance at half the cost,” and Cursor co-founder Sualeh Asif describes “near Fable 5 intelligence at Opus speed and cost.” One caveat from the release itself: Mythos 5 keeps the lead on offensive cybersecurity and biology research tasksand Anthropic's safety classifiers route flagged requests back to Opus 4.8.
What it means for the vendor map
Read the effort toggle as a pricing-power move, because that is what it is. Anthropic has been selling capability tiers as separate products, Haiku, Sonnet, Opus, Fable, each with its own price card. Opus 5 collapses a large stretch of that ladder into one SKU with a dial. The vendor now captures cost-sensitive workloads at low effort, mid-market workloads at medium, and frontier-adjacent workloads at high effort, all inside the same contract and the same integration. That structure defends revenue in both directions: it undercuts cheaper rivals before they peel off volume, and it self-cannibalizes Fable 5 demand on Anthropic's own terms rather than a competitor's.
For the rest of the vendor map, the pressure is direct. Any lab selling a frontier model at premium prices must now explain what the premium buys when a $5/$25 model sits within 0.5% of flagship scores on agentic-coding benchmarks. Expect rivals to answer within the quarter with effort-style controls, per-request pricing granularity, or straight price cuts. Cloud marketplaces reselling Claude inherit the toggle too, which turns every listing into a price-performance curve rather than a single line item. For buyers, the strategic shift is that model selection is becoming a runtime parameter rather than a procurement decision, which moves negotiating leverage toward whoever controls the routing layer. Enterprises that own their routing keep that leverage; enterprises hard-wired to a single vendor endpoint hand it back.
The 90-day decision
The concrete move for the next quarter: run a structured bake-off before your renewal window closes. Route a representative slice, 10% is enough, of current Opus 4.8 and Fable 5 traffic through Opus 5 at each effort level, and measure three things per workload class: task pass rate, tokens consumed per completed task, and end-to-end latency. Zapier-style workflow automations and Box-style document analysis are the workload types with vendor-verified gains, so start there. Teams holding Fable 5 committed-spend contracts should take the results into repricing conversations immediately; a vendor that just priced near-flagship capability at half the flagship rate has conceded the argument that frontier pricing reflected capability. Set effort defaults per workload class, low for routine extraction and classification, high for agentic coding, due diligence and long-running tasks, and wire the toggle into your gateway configuration so the dial stays in your hands, on your side of the contract.
Article by NOVAIndustry & Products
NOVA covers AI product launches and competitive moves for enterprise decision-makers.