Claude Opus 5 vs Fable 5: Six Benchmarks, Side-by-Side Pricing, and When to Switch
Opus 5 beats or ties Fable 5 on every published benchmark at half the API price - here is what the data actually says.
Choosing between Claude Opus 5 vs Fable 5 is the question most developers are asking after Anthropic's July 24 launch. On five of six benchmarks Anthropic published with the announcement, Opus 5 leads or ties Fable 5 while costing $5/$25 per million tokens - exactly half of what Fable 5 costs. One area still favors Fable 5. Most teams will never hit it.
Opus 5 Costs $5/$25 per Million Tokens - Fable 5 Charges Double
Pricing separates the two models immediately. Fable 5 runs at $10 per million input tokens and $50 per million output tokens; Opus 5 runs at $5 and $25 - the same as Opus 4.8. Cache reads cost $1.00 per million tokens on Fable 5 and $0.50 on Opus 5. A full breakdown of every Claude plan and API tier has the context on where these rates fit across the complete model stack.
| Model | Input ($/MTok) | Output ($/MTok) | Cache Read ($/MTok) |
|---|---|---|---|
| Fable 5 | $10 | $50 | $1.00 |
| Opus 5 | $5 | $25 | $0.50 |
| Opus 5 (Fast mode) | $10 | $50 | $1.00 |
Opus 5 Fast mode runs at roughly 2.5 times the default speed at twice the base price. Same per-token rate as Fable 5 standard. For teams choosing between them at that price point, Opus 5 leads on every published coding and agent benchmark - which makes the case for paying Fable 5's rate significantly harder to construct.
Benchmark by Benchmark: Opus 5 Leads on Five of Six
Anthropic published results across six evaluations with the Opus 5 launch - the full announcement is in our Claude Opus 5 launch coverage. Opus 5 takes the top spot on Frontier-Bench v0.1, Anthropic's internal software engineering benchmark, more than doubling Opus 4.8's score and surpassing Fable 5. On OSWorld 2.0, Opus 5 beats Fable 5's best computer-use result at just over a third of the cost, and at minimum effort still outperforms every other model in the field.
| Benchmark | Opus 5 result | vs Fable 5 | Notes |
|---|---|---|---|
| CursorBench 3.2 | Within 0.5% of Fable 5 at max effort | Effectively tied | Opus 5 at half the cost per task |
| Frontier-Bench v0.1 | #1 overall, 2x Opus 4.8 | Opus 5 leads | Anthropic's coding eval |
| ARC-AGI 3 | 3x higher than next-best model | Opus 5 leads | Novel problem-solving |
| OSWorld 2.0 | Beats Fable 5's best at 1/3 of the cost | Opus 5 leads | Computer-use tasks |
| Zapier AutomationBench | ~1.5x next-best pass rate at same cost | Opus 5 leads | Business task completion |
| Cybersecurity (exploit) | Close to Mythos 5 on finding vulns | No direct comparison published | Fable 5's classifiers fire 85% more often |
ARC-AGI 3 is the sharpest result. Opus 5 scores three times higher than the next-best model on a benchmark built specifically to test reasoning on genuinely novel problems - not patterns from training data. On Zapier's AutomationBench, which measures whether a model can complete real business workflows from start to finish without human intervention, Opus 5 posts about 1.5 times the next competitor's pass rate at identical cost per task. For teams also comparing these models against GPT-5.6 and Gemini, AI API Pricing Compared 2026 has the per-token costs across every major provider side by side.
Cybersecurity Exploitation Is Where Fable 5 Retains an Edge
Fable 5 carries more cybersecurity exploitation capability than Opus 5. Most developers cannot access it though, since both models block penetration testing and exploit generation via safety classifiers - and Fable 5's classifiers intervene around 85% more often than Opus 5's. For the Claude Opus 5 vs Fable 5 decision in security-sensitive environments, Anthropic's Cyber Verification Program matters more than the base model comparison: CVP enterprises get a less-restricted Opus 5 build immediately, and the highest-capability exploit work routes to Mythos 5 regardless.
Beyond cybersecurity, Anthropic has not published benchmarks where Fable 5 outperforms Opus 5. Biology-related requests that Fable 5 previously blocked now route to Opus 5 rather than Opus 4.8, which Anthropic frames as an upgrade path rather than a demotion. Fable 5 remains available on the API, but after this launch it no longer holds the top position on Anthropic's own coding and knowledge-work leaderboards.
API Teams Should Switch to Opus 5 for Most Workloads
For the Claude Opus 5 vs Fable 5 decision at the API level, the math favors Opus 5. Switching from Fable 5 cuts token costs in half. On every published coding and agent benchmark except CursorBench - where the gap is just 0.5% - Opus 5 leads outright. Teams running agentic pipelines or automation workflows that process millions of tokens pay twice as much per token to get lower benchmark results on Fable 5. That is a hard trade to justify, and the honest read is that most production teams have been paying Fable 5 prices for Opus 5-level work.
Claude Max and Claude Pro subscribers need no changes. Opus 5 is now the Claude Max default. For developers on the API, the claude-opus-5 identifier is live today. Teams assessing the full model tier also benefit from the Claude Sonnet 5 vs Opus 4.8 benchmark breakdown, which covers the adjacent matchup that set the baseline Opus 5 is now leapfrogging.
Two Beta API Features Change How Agents Handle Classifier Hits
Anthropic shipped two API-level beta features alongside Opus 5 that matter independently of the model comparison. Mid-conversation tool changes let developers swap which tools Claude can call within a session without invalidating the prompt cache - a rebuild that previously forced a full context reset in many agentic setups. Automatic fallbacks route requests flagged by safety classifiers to the best available model rather than returning a blocked error, so production pipelines no longer fail silently when a classifier fires.
Anthropic has not set a Fable 5 deprecation date. OpenAI and Google have not published direct responses to Opus 5's Frontier-Bench and OSWorld results, so the competitive picture for now relies entirely on Anthropic's own benchmark data - which teams should factor in when drawing conclusions about where Opus 5 sits relative to GPT-5.6 Sol and Gemini 3.5.
