alphatechfinance
Try it: claude.ai
NEW
CLAUDE OPUS 5
Anthropic shipped its fourth model in under two months, and this one changes the math. Claude Opus 5 comes close to the frontier intelligence of Fable 5 at half the price — and costs exactly the same as the Opus 4.8 it replaces. Here is what is new, what the benchmarks actually say, and who should care.
- The headline: near-Fable 5 intelligence at half the price, and the same $5/$25 pricing as Opus 4.8. It is now the default on Claude Max and the strongest model on Claude Pro.
- Coding: 43.3% on Frontier-Bench v0.1 — more than double Opus 4.8 (18.7%) and above Fable 5 (33.7%). On CursorBench 3.2 it lands within 0.5% of Fable 5's peak at half the cost per task.
- Beyond code: new state of the art on GDPval-AA, roughly 3× the next-best model on ARC-AGI 3, and it beats Fable 5's best OSWorld 2.0 result at just over a third of the cost.
- Alignment: Anthropic's most aligned model to date — 2.3 on overall misaligned behavior, the lowest of its recent models, with the lowest rates of deceptive behavior.
- Fewer roadblocks: cyber classifiers fire about 85% less often than on Fable 5, and flagged requests fall back to Opus 4.8 rather than being refused.
What Claude Opus 5 actually is
Anthropic describes Claude Opus 5 as a thoughtful, proactive model that comes close to the frontier intelligence of Fable 5 at half the price. That framing matters, because Fable 5 is the expensive, safeguard-heavy Mythos-class model most people cannot justify running daily. Opus 5 is built for exactly that gap: near-frontier capability you can use every day without the price or the friction.
It arrived on July 24, 2026 — only two months after Opus 4.8, and the fourth Claude model in under two months after Fable 5, Mythos 5, and Sonnet 5. It is now the default model on Claude Max and the strongest model available on Claude Pro, with API access through claude-opus-5, plus Amazon Bedrock, Google Cloud, and Microsoft Foundry.
Claude Opus 5 benchmarks, decoded
The benchmark story is not incremental. On several evaluations Opus 5 does not merely beat its predecessor — it reorders the leaderboard.
Frontier-Bench v0.1 measures whether a model can build working software from engineering drawings. Opus 5 scores 43.3%, more than doubling Opus 4.8's 18.7% and clearing Fable 5's 33.7% — at a lower cost per task. On CursorBench 3.2 at max effort, it performs within 0.5% of Fable 5's peak score for half the cost per task.
These are vendor-run evaluations on bounded tasks with clear outcomes. They are useful for ranking models against each other, not a substitute for testing on your own messy, real-world work. Treat the charts as a starting point.
What is new in Claude Opus 5
Benchmarks tell you the score. These are the behavioral changes that show up in daily use.
It verifies its own work
The clearest shift. Given a drawing to rebuild as a 3D CAD model — with no way to view the image — Opus 5 wrote its own computer vision pipeline to extract the geometry from raw pixels, then reconstructed the part. No competing model solved it in five attempts.
Effort settings and Fast mode
An effort dial trades intelligence for speed and token savings. Fast mode runs about 2.5 times the default speed at twice the base price, available on the Claude Platform and via usage credits in Claude Code.
Stronger science
It beats Opus 4.8 on every life sciences evaluation Anthropic tracks — 10.2 points higher on organic chemistry tasks like inferring structures from spectroscopy, and 7.7 points higher on protein-related work.
Better visual output
Anthropic demoed Opus 5 visualizing airflow over aerodynamic objects and building an interactive cell illustration. Early testers singled out the jump in 3D and front-end generation quality.
Two API updates in beta
Mid-conversation tool changes let developers swap available tools without invalidating the prompt cache. Automatic fallbacks route classifier-flagged requests to another model instead of blocking them.
Real efficiency gains
Customer reports back the efficiency claim: one trading benchmark hit its best Opus result using roughly a seventh of the reasoning tokens and under half the latency of Opus 4.8.
Claude Opus 5 alignment and safeguards
Anthropic's automated behavioral audit found Opus 5 to be its most aligned model to date. It scores 2.3 on overall misaligned behavior — the lowest of its recent models — adheres to Claude's Constitution better than Opus 4.8, Sonnet 5, or Fable 5, shows the lowest rates of deceptive behavior, and is the least susceptible to being tricked into misuse.
On dual-use risk, Opus 5 does not advance the frontier. It remains behind Mythos 5 in both biology research and offensive cybersecurity. Anthropic intentionally avoided training it on cyber tasks; it has still improved as a side effect of general capability, coming close to Mythos 5 at finding vulnerabilities while staying substantially behind on exploiting them — the step that turns a flaw into a real threat.
Practically, the guardrails are lighter than Fable 5's. Opus 5 can search source code for security issues but blocks binary-based vulnerability scanning, penetration testing, and exploit generation. Anthropic expects its classifiers to intervene around 85% less often than on Fable 5, and flagged requests in Claude.ai, Claude Code, and Cowork fall back to Opus 4.8 by default rather than hitting a refusal. Biology requests blocked on Fable 5 now route to Opus 5.
Who should switch to Claude Opus 5
If you were choosing between paying Fable 5 prices or accepting Opus 4.8 limits, that dilemma mostly disappears. Opus 5 delivers the majority of frontier capability at Opus economics, with lighter guardrails and no data retention requirements for general access.
Reach for Opus 5
- Long-running agents and multi-step coding work
- Knowledge work, research, and professional analysis
- Anything where cost per task actually matters
- Daily driver use on Max or Pro
Still reach for Fable 5
- When you need the absolute ceiling, cost aside
- Certain cybersecurity evaluations where Mythos leads
- Tasks where Fable one-shots what Opus needs retries for
- Long-horizon autonomous biology research
Benchmarks only go so far. Anthropic's official Claude channel publishes launch walkthroughs, demos, and deep dives that show how Opus 5 behaves on real tasks — worth ten minutes before you rebuild your workflow around it.
Claude on YouTubeOfficial demos, model launches, and technical deep dives Watch →Claude Opus 5 — quick answers
What is Claude Opus 5?
Anthropic's newest Opus-class model, released July 24, 2026. It approaches Fable 5's frontier intelligence at half the price, and is the default on Claude Max and the strongest model on Claude Pro.
How much does it cost?
$5 per million input tokens and $25 per million output tokens — the same as Opus 4.8. Fast mode runs about 2.5 times default speed at twice the base price.
How does it score on benchmarks?
43.3% on Frontier-Bench v0.1 (vs 18.7% for Opus 4.8 and 33.7% for Fable 5), within 0.5% of Fable 5's CursorBench peak at half the cost, plus leads on ARC-AGI 3, GDPval-AA, OSWorld 2.0, and AutomationBench.
Is it safer than previous models?
Anthropic calls it its most aligned model to date, scoring 2.3 on overall misaligned behavior. It stays behind Mythos 5 on biology and offensive cyber, and its cyber classifiers fire about 85% less often than Fable 5's.
Where can I use it?
On Claude Max and Pro, through the Claude API as claude-opus-5, and via Amazon Bedrock, Google Cloud, and Microsoft Foundry.
If this saved you a scroll, fuel the next one
Alpha Tech Finance covers AI and finance independently — researched, sourced, and free. A small contribution keeps it that way.
