On July 24, 2026, Anthropic launched Claude Opus 5, the new version of its most advanced AI model for complex tasks and autonomous agents. The promise: performance close to its most capable model, Fable 5, at half its price, and at the same price as the older Opus 4.8. For an SME leader who already uses Claude, ChatGPT, or Copilot day to day, this announcement confirms a broader trend: the most capable AI keeps getting more affordable, month after month. Here is what actually changes, with figures and sources.
In brief
- Claude Opus 5 launched on July 24, 2026, available on Claude.ai, the API, Claude Code, Claude Cowork, Amazon Bedrock, Google Cloud, and Microsoft Foundry (source: Anthropic).
- Price unchanged from Opus 4.8: $5 per million input tokens, $25 per million output tokens.
- Context window of 1 million tokens, with up to 128,000 output tokens.
- On the Frontier-Bench v0.1 benchmark, Opus 5 scores 43.3%, versus 33.7% for Fable 5 and 18.7% for Opus 4.8 (source: Anthropic).
- Becomes the default model on the Claude Max plan and the strongest model available on Claude Pro.
- For an SME, the real stakes are not chasing raw power but picking the right model for the right task.
What is Claude Opus 5, exactly?
Claude Opus 5 is the new generation of Anthropic's "Opus" line, designed for long and complex tasks: multi-step autonomous agents, analysis of large document sets, code generation and debugging. It rounds out the Claude family alongside Sonnet 5 (general-purpose, fast, cost-efficient) and Fable 5 (the most capable model, also the most expensive).
What stands out in this launch: Anthropic is not charging more for more power. The price stays identical to the older Opus 4.8, while benchmark scores rise sharply. This continues a trend seen over the past several months: the cost of the most advanced AI keeps falling relative to its performance, or even improves at the same price.
Scores on the Frontier-Bench v0.1 benchmark, source: Anthropic, announcement of July 24, 2026.
Why this launch matters for an SME
An SME does not need to track every model release. But three parts of this announcement have a direct impact on an AI budget:
- The cost-performance ratio of autonomous agents is improving. On OSWorld 2.0, a test that measures an AI's ability to operate a computer like a human (clicking, filling forms, navigating between apps), Opus 5 beats Fable 5's best score at roughly a third of the cost per task (source: Anthropic).
- Customer feedback is concrete and measurable. Ben Kus, CTO of Box, reports an 8% overall accuracy gain compared with Opus 4.8, with a 17% gain on due-diligence document tasks. Wade Foster, CEO of Zapier, says Opus 5 topped his company's internal AutomationBench leaderboard, scoring 100% on a customer-churn prediction task no previous model had passed.
- Claude Sonnet 5's pricing schedule is also changing. Sonnet 5's introductory price ($2 / $10 per million tokens) ends on August 31, 2026; it then rises to $3 / $15. An SME building a high-volume automation on Sonnet 5 should plan for this change in its budget.
July 24, 2026
Claude Opus 5 launches
August 31, 2026
Sonnet 5 introductory pricing ends
Which model for which task?
The main risk for an SME is not missing a model release, but overpaying for a flagship model on simple tasks, or the opposite: under-equipping a project that needs reliability.
Claude Sonnet 5
Claude Opus 5
| Model | Input / output price (per million tokens) | Context | Frontier-Bench v0.1 score | Recommended use |
|---|---|---|---|---|
| Claude Opus 4.8 | $5 / $25 | 1M tokens | 18.7% | Older flagship model, now lagging |
| Claude Sonnet 5 | $2 / $10 (until 08/31/26), then $3 / $15 | - | - | Everyday, high-volume use |
| Claude Opus 5 | $5 / $25 | 1M tokens | 43.3% | Complex agents, code, document analysis |
| Claude Fable 5 | $10 / $50 | - | 33.7% | Most demanding cases, dedicated budget |
Sources: Anthropic (announcement of July 24, 2026), Claude Platform pricing documentation.
What to keep in mind
No model is best at everything. On DeepSWE v1.1, a software bug-fixing benchmark, an OpenAI competitor model (GPT-5.6 Sol) scores higher than Opus 5. On some legal benchmarks, Fable 5 still leads. Model choice should be made task by task, not on general reputation.
A stable price per token does not mean a stable bill. A multi-step autonomous agent can consume a lot of output tokens, up to 128,000 per request with Opus 5. Without a cap or usage tracking, the monthly cost can climb quickly.
Governance still matters. Who in the company has access to which model? What data can pass through the API? What is the monthly budget per team? These questions matter more than the choice of model itself.
Key takeaway
A more capable model at the same price does not justify migrating everything at once. Test Claude Opus 5 on one high-value task (due diligence, code review, complex automation) before rolling it out broadly, and keep Sonnet 5 for everyday volume.
FAQ
Is Claude Opus 5 more expensive than Claude Opus 4.8?
No. The price stays the same: $5 per million input tokens and $25 per million output tokens, for markedly higher performance on the benchmarks Anthropic published on July 24, 2026.
Should an SME migrate all its usage to Claude Opus 5?
No. For simple, high-volume tasks (routine writing, customer support, summaries), Claude Sonnet 5 remains more cost-efficient. Claude Opus 5 makes sense for complex tasks: multi-step autonomous agents, cross-document analysis, code generation.
What does the 1-million-token context window actually change?
It lets the AI process the equivalent of several hundred pages (contracts, reports, codebases) in a single pass, without manual splitting. Useful for due diligence or document audits, less useful for short exchanges.
Where can Claude Opus 5 be used?
On Claude.ai (Pro and Max plans), via the Anthropic API, in Claude Code and Claude Cowork, and via Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
Going further
Choosing the right AI model at the right price is an ongoing exercise in 2026. LUWAI has already covered Claude Fable 5's shift to usage-based billing and Google's Gemini 3.6 Flash price cut. Find all our analyses on the Resources page, or see how SMEs have structured their AI adoption in our customer success stories.
