Anthropic released Claude Opus 5 on July 24, 2026, at the same price as its predecessor but with significantly stronger performance, positioning it as the default choice for most coding teams instead of the more expensive Fable 5. The new model costs $5 per million input tokens and $25 per million output tokens, identical to Opus 4.8, yet scores 61 on the Artificial Analysis Intelligence Index, just one point below Fable 5 and two points above GPT-5.6 Sol. This pricing strategy represents a deliberate shift: Anthropic held the line on cost while lifting capability, making everyday AI work more affordable for teams running coding agents and automation pipelines.

The release marks Anthropic's fourth model launch in under two months, following Mythos, Fable, and Sonnet 5. This accelerated pace reflects the company's confidence in its ability to iterate rapidly on frontier models. What sets Opus 5 apart is not just raw performance, but the structural changes to how teams interact with it. The model introduces a five-step effort ladder, allowing engineers to dial intelligence up or down per request rather than switching between entirely different models. This means a simple data-extraction task can run at low effort for speed and cost, while a complex debugging session can call the same model at maximum effort without forcing the entire pipeline to a more expensive variant.

How Does Opus 5 Compare to Fable 5 on Real Coding Tasks?

On benchmarks that matter for software engineering, Opus 5 either matches or comes within striking distance of Fable 5. On CursorBench 3.2, a test that measures in-IDE coding assistance, Opus 5 at maximum effort lands within 0.5% of Fable 5's peak score while costing roughly half as much per completed task. On the Artificial Analysis Coding Agent Index, Opus 5 at maximum effort with Claude Code achieves joint first place, including the highest score on SWE-Atlas-QnA, a benchmark for software engineering question-answering. On Frontier-Bench v0.1, an agentic software engineering test, Opus 5 surpasses every other model and more than doubles Opus 4.8's score at a lower cost per task.

The cost-per-task metric reveals the real story. Opus 5 costs $2.03 per task on the Artificial Analysis Intelligence Index at maximum effort, compared to Fable 5's $2.75, a 26% savings. This advantage grows at higher effort levels, where Opus 5 wastes fewer tokens reaching correct answers than both Opus 4.8 and Sonnet 5. However, Fable 5 still holds narrow advantages in specific domains. It leads on factual knowledge benchmarks, where Opus 5 trails slightly and shows a 14-point increase in hallucination rate to 50%, pushing its accuracy down on retrieval-heavy or compliance-sensitive tasks. For teams building long-running autonomous agents that operate for days without human intervention, Fable 5 remains Anthropic's recommended choice.

What Are the Key Differences in How Opus 5 Works?

• Effort Ladder: Five settings (low, medium, high, xhigh, max) let you tune intelligence per request instead of switching models entirely, with output token usage ranging roughly 8x from low to maximum effort.
• Thinking Mode: Thinking is enabled by default and cannot be disabled at xhigh or maximum effort, a breaking change for automation that previously relied on thinking-off at high effort levels.
• Fast Mode: A new option runs the model at roughly 2.5x standard speed at twice the base price, $10 per million input tokens and $50 per million output tokens, for latency-sensitive workloads.
• Context Window: A 1-million-token context window allows processing roughly 100,000 words at once, with 128,000 maximum output tokens per response.
• Cache Pricing: Cache writes carry a 25% premium at $6.25 per million tokens with a 5-minute time to live, while cache hits receive a 90% discount at $0.50 per million tokens, allowing agents that reuse large codebases to cut real input costs well below the headline rate.

Where Does Opus 5 Stand Against OpenAI's GPT-5.6 Sol?

Opus 5 leads on 9 of 12 shared benchmarks against GPT-5.6 Sol, with particularly strong performance on SWE-bench Pro, where it holds a substantial lead, and ARC-AGI-3, a novel-reasoning test where Opus 5 scores roughly four times higher. Opus 5 also costs 17% less on output tokens, $25 versus $30 per million. However, Sol retains genuine advantages for specific use cases. It leads on Terminal-Bench 2.1 when using parallel sub-agents for terminal-based agentic coding, and its Cerebras deployment option delivers raw throughput around 750 tokens per second, more than ten times Opus 5's standard speed. For teams whose workflows are heavily terminal-based with a premium on raw latency, Sol remains a compelling alternative.

The broader pattern across all comparisons is consistent: Opus 5 either matches Fable 5 or gets close, and it does so for less money per task. The one coding-shaped area where it clearly is not the leader is offensive security, which Anthropic describes as a deliberate design choice. This positioning reflects Anthropic's intent for Opus 5 to be the everyday model, not the specialized frontier tool.

What Does This Mean for Teams Building AI Agents?

For most teams, the decision framework simplifies significantly. Use Opus 5 as the default for everyday engineering work: in-IDE assistance, pull-request review, bug fixing, multi-file refactors, and terminal or DevOps automation. Anthropic designed it to be used every day and made it the default model on Claude Max, a strong signal about where the company expects volume to concentrate. Reserve Fable 5 for the narrower set of tasks where three conditions hold true: factual precision is the product, the horizon is very long (agents running autonomously for days), or the task sits in Fable 5's remaining lead on frontier science.

Claude Sonnet 5 remains the value floor at $1.53 per Index task, the cheapest option in the lineup for high-volume, latency-sensitive, or budget-capped agents. The honest framing is a ladder: Sonnet 5 for volume, Opus 5 for quality at reasonable cost, and Fable 5 for the frontier cases where nothing else suffices. The effort ladder adds a new dimension to this routing logic. Instead of choosing a model per task, teams can now choose an effort level per step, keeping a single model in the pipeline and tuning cost and quality on the fly. This flexibility is particularly valuable for multi-step agent pipelines where different steps have different requirements.

One practical caveat deserves emphasis: the 26% cost savings is real only if teams also right-size the effort setting. Paying maximum-effort prices for medium-effort work defeats the purpose of the migration. Teams switching from Fable 5 to Opus 5 should audit their workloads, identify which steps genuinely need maximum effort, and dial down the rest. The same principle applies to thinking mode. Since thinking cannot be disabled at xhigh or maximum effort on Opus 5, any automation that assumed thinking-off was always available will need adjustment before switching models.

Anthropic's release framing is unusually direct for a frontier lab: Opus 5 is described as "a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price". This positioning reflects real market feedback. Fable 5 drew criticism over its token burn rate and cost, factors behind Anthropic's recent subscription pricing changes for Fable 5. Opus 5 looks like the direct response, the everyday model that most workloads should actually run on, with Fable 5 reserved for the cases that genuinely need it. For teams currently paying for Fable 5 purely because it was the best coding model, Opus 5 is the cheaper default that gives up very little.