Claude Opus 5 Matches Anthropic’s Priciest Model at Half the Token Cost
Anthropic's new Claude Opus 5 posts top coding and knowledge-work scores while approaching Fable 5's performance at half the token price, with a startling ARC-AGI-3 result.
In Brief
- Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens—half of Fable 5’s rates—while matching or beating it across most of Anthropic’s own benchmarks
- On ARC-AGI-3, which measures novel problem-solving, Opus 5 scores 30.2 percent, nearly four times higher than GPT-5.6 Sol’s 7.8 percent and far above Opus 4.8’s 1.5 percent
- Opus 5 becomes the default model on Claude Max and the most capable model on Claude Pro, with early independent benchmarks backing Anthropic’s price-performance claims
Anthropic’s new flagship model, Claude Opus 5, posts top scores in coding and knowledge work while approaching the performance of the much pricier Claude Fable 5 at half its token rates. The launch is a direct response to pricing pressure from GPT-5.6 Sol and Chinese competitors, and it reshapes the value equation for anyone building on frontier models, The Decoder reported.
Opus 5 becomes the default model on Claude Max and the most capable option available on Claude Pro. Its 1 million-token context window and token rates are unchanged from its predecessor: like Opus 4.8, it costs $5 per million input tokens and $25 per million output tokens. A new Fast Mode increases speed by 2.5x but doubles the price.
The pricing gap with Fable 5 is stark. Fable 5 runs $10 per million input tokens and $50 per million output tokens — exactly double Opus 5 across the board. Anthropic’s pitch is that customers no longer have to pay Fable 5 rates to get near-Fable 5 results.
Where Claude Opus 5 wins and where it doesn’t
According to Anthropic’s own benchmarks, Opus 5 sets records across several evaluations. On Frontier-Bench v0.1, it hits 43.3 percent on agentic terminal coding, beating Fable 5 (33.7 percent), GPT-5.6 Sol (34.4 percent), and its predecessor Opus 4.8 (21.1 percent) by wide margins. On the GDPval-AA v2 knowledge-work benchmark, Opus 5 leads with an Elo of 1,861, ahead of Fable 5 (1,747) and GPT-5.6 Sol (1,736).
It does not win everywhere. On agentic coding via DeepSWE v1.1, GPT-5.6 Sol leads with 72.7 percent, followed by Fable 5 (69.7 percent) and Opus 5 (68.8 percent). Opus 5 also falls behind Anthropic’s own Mythos 5 on cybersecurity tasks — the company says it deliberately did not train Opus 5 on cyber work, as with its predecessor.
The biggest surprise is ARC-AGI-3, a benchmark that measures novel problem-solving without memorized patterns. Opus 5 scores 30.2 percent, roughly four times GPT-5.6 Sol’s 7.8 percent, while Opus 4.8 managed just 1.5 percent. There is no Fable 5 result for the test, and Anthropic notes it is unclear whether such a large lead will show up in everyday use.
Effort settings and the token-efficiency catch
Users can trade performance against token use through five effort settings — low, medium, high, xhigh, and max. In its prompting guide, Anthropic recommends broad use of “low” and “medium” for good results at a fraction of the token cost and latency, while suggesting developers start with “xhigh” for coding and agentic tasks.
Token rates alone do not tell the full story, because token efficiency varies. Opus 4.7 ended up costing 30 to 40 percent more per task than Opus 4.6 despite identical base rates. Curiously, Opus 5 scores slightly worse at the max effort setting than at the second-highest setting on two benchmarks despite costing more; a FrontierCode developer explained that at higher effort Opus 5 tends to refactor surrounding code even when a small fix would do, and the benchmark counts those unsolicited changes as errors.
Anthropic also says Opus 5 can check and improve its own work through iteration and build its own tools through code when it needs them. Early independent benchmarks back the company’s claims: Opus 5 costs less than Fable 5 while often performing better, and it also beats both Opus 4.8 and Sonnet 5 on price and performance.
FAQ
How much cheaper is Claude Opus 5 than Fable 5?
Opus 5 costs $5 per million input tokens and $25 per million output tokens, exactly half of Fable 5’s $10 and $50 rates. Anthropic positions it as delivering near-Fable 5 performance at half the token price.
What is the most striking Claude Opus 5 benchmark result?
On ARC-AGI-3, a test of novel problem-solving, Opus 5 scores 30.2 percent — nearly four times GPT-5.6 Sol’s 7.8 percent and far above Opus 4.8’s 1.5 percent. There is no Fable 5 score for this benchmark.
Does Claude Opus 5 beat every rival model?
No. GPT-5.6 Sol leads on DeepSWE v1.1 agentic coding at 72.7 percent, and Anthropic’s own Mythos 5 outperforms Opus 5 on cybersecurity tasks because Anthropic deliberately did not train Opus 5 on cyber work.
