Last Updated: 27 July 2026
Anthropic released Claude Opus 5 on July 24, 2026, describing it as a model built for complex coding, enterprise work, research and everyday professional tasks. The model became the default option on Claude Max and the strongest Opus model offered on Claude Pro as soon as it launched, putting it in front of a large share of Anthropic's paying customers immediately.
The most relevant detail is price. Opus 5 keeps the same API rate as its predecessor, Opus 4.8, at $5 per million input tokens and $25 per million output tokens. Claude Fable 5, Anthropic's most capable widely released model, costs twice as much at $10 and $50. That gap sits at the center of how Anthropic wants developers and businesses to think about the two models.
That framing turns the Opus 5 launch into a value question rather than a simple upgrade story. The rest of this article looks at what that value actually covers, where Opus 5 genuinely closes the gap with Fable 5, and where Fable 5's higher price still buys something Opus 5 cannot match.
Anthropic's model documentation places Opus 5 as the option for complex agentic coding and enterprise work, while Fable 5 is described as next-generation intelligence built for agents that run over long stretches of time. Both models share the same 1-million-token context window and the same maximum synchronous output of 128,000 tokens, so context size is not what separates them.
Latency differs more clearly. Anthropic lists Opus 5 with moderate comparative latency and Fable 5 as the slower of the two. Both models are reachable through the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud and Microsoft Foundry, using the model IDs claude-opus-5 and claude-fable-5.
Reasoning style differs too. Fable 5 keeps adaptive thinking switched on constantly. Opus 5 defaults to the same approach yet gives engineers more room to scale it back at lower effort levels. Anthropic lists a reliable cutoff of May 2026 for Opus 5 against January 2026 for Fable 5. That gap informs planning. It does not prove either system holds a wider edge overall. Each tool still targets a different job.
At standard rates, Opus 5 runs at half the price of Fable 5. That gap holds steady across the board. Prompt caching keeps the same ratio, and Anthropic's Batch API applies its usual discount evenly to both models, so the price difference between them never narrows or widens depending on how the tokens are billed.
Opus 5 also offers a Fast mode, still a research preview limited to Anthropic's own API and Managed Agents. It trades a speed boost, well over double the normal output pace, for pricing that lines up with Fable 5's standard rate. Anthropic clarifies that this extra speed shows up in how quickly the response is generated, not in how fast it starts.
Token price is not the same thing as the total cost of getting a task done. Effort level, the number of tokens generated, cache use, tool calls, failed attempts and how much supervision a task needs can all push the final cost of a successful result well above or below the sticker price. Anthropic's own comparison between the two models on CursorBench refers to cost per completed task rather than the raw per-token rate.
A lower published price is easy to verify, but it only tells part of the story. Whether Opus 5 actually costs less for a given piece of work depends on how many attempts and how much output that work requires, which is exactly the question the next section examines through Anthropic's benchmark results.
Cursor's CursorBench 3.2 evaluates coding agents on ambiguous, multi-file tasks pulled from real Cursor sessions, testing instruction following and tool use under realistic conditions. On the leaderboard's live results, Fable 5 Max scores 70.5% with an average cost of $17.32 per task, while Opus 5 Max scores 70.0% at an average cost of $8.23. That is a gap of half a percentage point in score, against a cost roughly half as high for Opus 5.
Anthropic describes the outcome as nearly matching Fable 5, yet the two scores are close enough that the more honest way to put it is a slim gap on a single benchmark. Cursor explains that its cost figures come from applying published pricing to its own set of tasks, and cautions that such tight score differences may not mean much statistically.
Anthropic shares further findings beyond this one test. On its internal Frontier-Bench run, the company reports that Opus 5 roughly doubled the performance of Opus 4.8 while spending less per task, although that number reflects a harness that tries several times and reverts to the older model when safety checks reject an answer.
Taken together, the benchmark picture supports Anthropic's cost argument more clearly than it supports a claim of equal capability. Opus 5 is landing close to Fable 5 on specific, narrow tests built around coding and computer use, at a meaningfully lower cost per task. Whether that gap holds on a wider range of everyday work is a separate question the published benchmarks do not answer.
Anthropic continues to describe Fable 5 as its most capable widely released model, and its main selling point is not a bigger context window but staying power. Fable 5 can work inside an agent harness for days at a time, planning across stages, delegating to subagents and checking its own output as it goes. Anthropic points to large code migrations, complex implementations and demanding research as the use cases where this matters most.
That framing lines up with what Dianne Penn told Reuters: choose Fable 5 for highly autonomous, multi-day projects, and choose Opus 5 when value is the priority. Fable 5's always-on adaptive thinking also means developers have less room to dial down reasoning to cut cost, though effort settings remain available. Fable 5 additionally supports Anthropic's Priority Tier, an option Opus 5 does not offer, which may matter to enterprise customers with committed capacity arrangements.
None of the available sources offer a broad, independent measurement of how much further Fable 5 gets on a genuinely long, multi-day project compared with Opus 5. The case for Fable 5 rests more on how Anthropic positions the model than on a published head-to-head test of task completion over days rather than minutes.
Even so, the logic behind paying more for Fable 5 is straightforward in cases where failure is expensive. If a long autonomous run fails partway through, or a missed requirement forces a costly restart, the extra token spend on Fable 5 can be easier to justify than it would be for routine, short tasks. That is an editorial reading of how the two models are positioned, not a measured result.
Anthropic's own documentation recommends starting with Opus 5 for complex agentic coding and enterprise work, and moving to Fable 5 only when a task genuinely needs the highest available capability. In practice, Opus 5 looks like the stronger default for frequent coding, debugging, document work, research and automation where controlling API spend matters. Fable 5 is worth testing specifically for large migrations, multi-stage agents and work expected to run for extended periods without close supervision.
How each model handles data can decide things by itself. The larger models hold onto information for 30 days and stay off-limits under strict zero-retention setups unless a business enables that separately, while the lighter option has no such restriction for typical use. This mainly touches bigger organizations with tighter privacy needs, not casual users. From there, it helps to judge cost per successful outcome instead of comparing scores alone.
The pricing and positioning may suggest that Anthropic is turning advanced agent capability into a model built for regular paid use, while keeping Fable 5 reserved for costlier, more autonomous and more sensitive work. That pattern could raise the question of whether Anthropic's model lineup is increasingly shaped by operating cost and safety controls as well as raw capability, though Anthropic has not stated this as company strategy.
Opus 5 was not designed to improve at cyberattacks. Even so, its broader gains ended up carrying over into some security work. The model does a solid job finding weak spots in code. It still struggles to build attacks that actually succeed. That is why certain penetration tests and exploit creation stay off limits.
Anthropic notes that its safety filters kick in much less often for this model compared to the more advanced one. That does not mean less real risk. It just means fewer stops along the way. In everyday apps, a blocked request automatically shifts to an older model. Certain approved teams can use a version with lighter restrictions.
The cybersecurity comparison is a reminder that Opus 5 and Fable 5 differ in more than price and speed. Their safeguards are tuned differently, and understanding those differences matters for any organization planning to rely on either model for security-adjacent work.
Opus 5 is now the default model on the top consumer plan and the strongest option on the mid-tier plan, available through Anthropic's usual developer channels and major cloud providers. Fable 5 comes with every paid plan, though not the free tier, and its current access rules have been in place for about a week.
On higher-tier plans, Fable 5 draws from a portion of the normal weekly usage limit and burns through that shared allowance faster than other models. On lower-tier plans, it instead pulls from pay-as-you-go credits right away rather than any included allowance. Fable 5 works across most of Claude's apps, though the coding tool needs a recent version to support it.
Opus 5 also includes a faster mode, separate from regular plan access. It remains an early, developer-only preview with its own pricing and limits, not a speed option available everywhere. Fallback behavior differs too: everyday apps quietly switch to an older model when a request gets blocked, while developers can choose how that fallback works themselves.
Anyone deciding between the two models should check current plan entitlements before committing, since usage rules and access conditions for newly released models tend to shift in the weeks after launch.
Claude Opus 5 is Anthropic’s new high-end model for coding, research, automation and professional work. It is positioned below Fable 5 in Anthropic’s model lineup, but the company says it delivers similar results on selected tests at a lower cost. Opus 5 is also the default model for Claude Max and the strongest model available on Claude Pro. Its main appeal is value, giving developers and businesses access to advanced capabilities without paying the higher API rates charged for Fable 5.
Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens through the standard API. Fable 5 costs $10 per million input tokens and $50 per million output tokens, making Opus 5 exactly half the standard token price. However, total task cost also depends on effort level, token use, retries and completion quality. Opus 5 Fast mode costs twice the normal Opus 5 rate, bringing its listed token price to the same level as standard Fable 5.
The two models can perform similarly on some tasks, but Fable 5 remains the higher capability tier. Anthropic says Opus 5 nearly matched Fable 5 on a CursorBench coding evaluation and performed strongly on other computer-use tests. These company-reported results do not establish that the models are equal in research, business work or long autonomous projects. Fable 5 may still offer an advantage when a task requires extensive planning, delegation, self-checking or several stages of work without frequent human guidance.
Developers and businesses should consider Opus 5 first for regular coding, debugging, analysis, research and automation. Its lower API price makes it easier to use across frequent or high-volume workloads. Fable 5 is more suitable when a project is unusually complex, may run autonomously for days or would be expensive to restart after failure. The best choice depends on total completion cost, required supervision, latency, data-retention rules and reliability on the exact task, not only on benchmark scores or token prices.
Claude Opus 5 is available to paid Claude users and developers through Anthropic’s platforms. It can also be used through supported cloud providers, although availability may differ by service. Claude Max receives Opus 5 as its default model, and Claude Pro users can select it as the strongest Opus option. Developers should note that Fast mode has separate pricing and access conditions. They should also monitor fallback behavior because restricted requests may be completed by another model rather than Opus 5 itself.
