Anthropic's Claude Opus 5 hits near-parity with Fable 5 at half the price
The new model is state-of-the-art on coding and knowledge-work benchmarks while costing the same as its predecessor and avoiding Fable 5's conservative guardrails.
What matters
- Anthropic released Claude Opus 5 on July 24, 2026, claiming near-parity with Fable 5 at half the price ($5/$25 per million input/output tokens, same as Opus 4.8).
- Opus 5 is state-of-the-art on Frontier-Bench v0.1 and GDPval-AA, within 0.5% of Fable 5 on CursorBench 3.2 at max effort, scores 30.2% on ARC-AGI 3 (~4x GPT-5.6 Sol), and beats Fable 5 on OSWorld at ~1/3 the cost.
- The model includes a 1M-token context window, an adjustable effort setting, a new Fast Mode (2.5x speed at double the price), prompt-cache-friendly tool changes, and token-efficiency improvements.
- Opus 5 is less restrictive than Fable 5, whose conservative safeguards triggered in ~5% of sessions and sometimes caught harmless requests; Fable 5 was also placed under strict US government export controls.
- Anthropic's model ladder runs Haiku 4.5 ($1), Sonnet 5 ($2 introductory, expiring Aug 31), Opus 5 ($5), Fable 5 ($10), with Mythos 5 not sold publicly; the launch comes amid pricing pressure from GPT-5.6 Sol and Chinese competitors.
Launch facts
- Price:
- $5 per million input tokens, $25 per million output tokens (same as Opus 4.8); Fast Mode doubles price for 2.5x speed
- Availability:
- Available July 24, 2026 across Claude.ai, Claude API, Claude Code, Claude Cowork, and major cloud platforms; default on Claude Max, strongest on Claude Pro
- Platforms:
- Claude.ai, Claude API, Claude Code, Claude Cowork, major cloud platforms
Benchmarks
| Benchmark | Claude Opus 5 | Claude Opus 4.8 | Claude Fable 5 | GPT-5.6 Sol | Source |
|---|---|---|---|---|---|
| Frontier-Bench v0.1 | state-of-the-art (surpasses all other models; more than doubles Opus 4.8) | baseline (less than half of Opus 5) | — | — | vendor-reported |
| CursorBench 3.2 (max effort) | within 0.5% of Fable 5 peak | — | peak score (reference) | — | vendor-reported |
| ARC-AGI 3 | 30.2% | — | — | ~7.5% (Opus 5 is ~4x higher) | vendor-reported |
| OSWorld (computer-use) | beats Fable 5 best result at ~1/3 cost | — | — | — | vendor-reported |
| GDPval-AA | state-of-the-art | — | — | — | vendor-reported |
Numbers come from the linked sources; vendor-reported results are labeled and worth independent verification.
What happened
Anthropic released Claude Opus 5 on July 24, 2026, positioning it as an everyday workhorse that comes close to the frontier intelligence of its pricier flagship, Claude Fable 5, at half the token cost. Opus 5 is priced at $5 per million input tokens and $25 per million output tokens — identical to its predecessor, Opus 4.8, and half of Fable 5's $10/$50 rates. It is available immediately across Claude.ai, the Claude API, Claude Code, Claude Cowork, and major cloud platforms, and is now the default model on Claude Max and the strongest option on Claude Pro.
On Anthropic's own coding and knowledge-work evaluations, Opus 5 is state-of-the-art on Frontier-Bench v0.1 and GDPval-AA. It more than doubles Opus 4.8's score on Frontier-Bench at a lower cost per task, and on CursorBench 3.2 at max effort it lands within 0.5% of Fable 5's peak score at roughly half the cost per task. On ARC-AGI 3, a benchmark built around novel problem-solving, Opus 5 scores 30.2% — nearly four times higher than GPT-5.6 Sol, according to Anthropic. On the OSWorld computer-use benchmark, the company says Opus 5 beats Fable 5's best result at just over a third of the cost. Opus 5 remains behind Anthropic's restricted Mythos 5 model on cybersecurity tasks.
The model includes a 1M-token context window, an adjustable effort setting that lets developers trade reasoning depth for lower token cost and faster responses, and a new Fast Mode that increases speed by 2.5x at double the price. Anthropic also says Opus 5 uses fewer tokens for similar quality and introduces prompt-cache-friendly tool changes, which could reduce costs further beyond the headline per-token price.
Notably, Opus 5 is less restrictive than Fable 5. Fable 5 — which was placed under strict US government export controls — launched with conservatively tuned safeguards that triggered in less than 5% of sessions but sometimes caught harmless requests. TechCrunch notes that Opus 5 will be both cheaper and less restrictive than Fable, likely making it preferable in most use cases.
For context, Anthropic's model lineup now runs: Haiku 4.5 at $1 per million input tokens, Sonnet 5 at $2 on introductory pricing expiring August 31, Opus 5 at $5, and Fable 5 at $10. Above all of them sits Mythos 5, which Anthropic does not sell to the public because of its cybersecurity capabilities.
Why it matters
The launch is Anthropic's fourth model release in under two months, and it comes amid pricing pressure from OpenAI's GPT-5.6 Sol and Chinese competitors. By offering near-frontier performance at Opus-tier pricing, Anthropic is making a direct play for developers and enterprises that want Fable 5-class capability without the cost or the friction of conservative guardrails. The effort-setting feature and Fast Mode give teams more knobs to tune cost and latency for their specific workloads, while the prompt-cache-friendly tool changes and token-efficiency claims could compound the savings.
The framing is also a shift for Anthropic: rather than leaning on beating rival labs, this launch leans on beating its own top-tier model on price — a signal that the competitive frontier is moving from raw capability to price-performance and usability.
What to watch
- Whether Opus 5's token-efficiency and prompt-cache improvements translate to real-world cost savings beyond the headline per-token price cut.
- How quickly developers migrate Fable 5 workloads to Opus 5, and whether Fable 5's export-control restrictions accelerate that shift.
- Whether Anthropic's pricing pressure from GPT-5.6 Sol and Chinese competitors leads to further price cuts or new tier introductions.
- Whether Opus 5's reduced restrictiveness holds up safely in production, or whether Anthropic tightens guardrails in response to incidents.
What to do next
Developers
Route a representative sample of production prompts to Opus 5 at default, high, and max effort settings, and compare cost, token consumption, refusal rates, and output quality against Fable 5. Also test Fast Mode for latency-sensitive workloads and measure prompt-cache hit rates.
Opus 5 is within 0.5% of Fable 5 on CursorBench at half the cost, uses fewer tokens for similar quality, and introduces prompt-cache-friendly tool changes — empirical testing can confirm the right configuration for your workload.
Founders
Switch your default Anthropic model to Opus 5 for features that don't require Fable 5's frontier capabilities, and estimate the impact on unit economics using the effort-setting feature and token-efficiency improvements.
At half the token cost of Fable 5 with near-parity on coding and knowledge work, plus token-efficiency gains, Opus 5 could meaningfully reduce per-user inference costs without an obvious capability downgrade.
PMs
Audit current product features that rely on Fable 5 and identify where Opus 5's fewer restrictions, lower cost, 1M-token context window, and prompt-cache-friendly tool changes would improve user experience without compromising safety requirements.
Fable 5's conservative safeguards triggered in ~5% of sessions and sometimes caught harmless requests; Opus 5's reduced restrictiveness, larger context window, and tool-change improvements may reduce friction in user-facing features.
Investors
Monitor Anthropic's API revenue signals and enterprise adoption metrics following the Opus 5 launch, particularly whether the lower price point and token-efficiency claims accelerate usage volume against GPT-5.6 Sol and Chinese competitors.
A near-frontier model at half the cost of the flagship, with efficiency improvements, could drive volume-based revenue growth and market share gains, but margin compression and competitive pricing pressure are worth watching.
Operators
Run a cost simulation comparing Opus 5 and Fable 5 across your current monthly token volume, factoring in the effort setting, Fast Mode, prompt-cache hit rates, and token-efficiency gains to model best-case and worst-case spend.
Opus 5's half-price positioning, adjustable effort settings, Fast Mode (2.5x speed at double the price), prompt-cache-friendly tool changes, and token-efficiency claims create multiple cost scenarios; quantifying potential savings requires modeling against actual usage patterns.
How to test
- 1Check the Anthropic API documentation and pricing page for Opus 5 availability and per-token costs, including Fast Mode pricing.
- 2Route a sample of production prompts to Opus 5 at default, high, and max effort settings, logging responses, token consumption, latencies, and any refusals.
- 3Run the same prompt set through Fable 5 to create a baseline for cost, quality, token usage, and restriction behavior.
- 4Test Fast Mode separately on latency-sensitive prompts, measuring speedup and cost-per-request against standard Opus 5 and Fable 5.
- 5Measure prompt-cache hit rates and cache-related cost savings for both models, especially around tool-change scenarios where Opus 5's prompt-cache-friendly changes should help.
- 6Test long-context prompts approaching the 1M-token context window to verify performance and cost at scale.
- 7Compare total token costs, refusal rates, token consumption, and output quality between the two models across effort settings.
- 8Scale the cost comparison to your monthly token volume to estimate savings, and identify the effort setting that best balances cost and quality for your workload.
Caveats
- Opus 5's exact per-token pricing is derived from Anthropic's 'half of Fable 5' claim and third-party Fable 5 pricing data; confirm directly with Anthropic's pricing page before modeling.
- The meaning of 'less restrictive' is not fully detailed in available sources; verify that Opus 5's safety policies meet your compliance requirements.
- Opus 5 remains behind Mythos 5 on cybersecurity tasks; do not assume parity across all domains.
- Benchmark results are vendor-reported and should be validated against your own workload.
- Fast Mode doubles the price for 2.5x speed; cost-effectiveness depends on your latency requirements.
- Token-efficiency and prompt-cache-friendly tool change claims are vendor-reported; measure actual savings on your workload.