Anthropic Launches Claude Opus 5: Near-Frontier Intelligence at Opus 4.8 Cost
On July 24, 2026, Anthropic officially launched Claude Opus 5. Its core selling point is straightforward: pricing identical to the previous Opus 4.8 ($5 per million input tokens, $25 per million output tokens), but with substantially stronger performance that sets new industry records on several coding and knowledge work benchmarks, with overall capability approaching the higher-tier Claude Fable 5 — in other words, you can get near-frontier intelligence at roughly half the price. For teams selecting models for websites, SaaS products, or agent applications, this is an option worth serious evaluation.
Performance Leap at the Same Cost
Opus 5's progress is most visible in software engineering tasks. On Frontier-Bench v0.1, Opus 5 surpasses all other models, with per-task cost below half of Opus 4.8 while more than doubling its performance. On CursorBench 3.2, at max effort, it reaches 99.5% of Fable 5's peak score at half the cost per task.
Knowledge work is equally impressive:
- On ARC-AGI 3, which tests solving novel problems, Opus 5's score is three times that of the next-best model;
- On Zapier AutomationBench, which measures end-to-end business task completion, its pass rate is about 1.5x the next-best model at the same cost per task — even at its lowest effort setting it completes more business tasks than any other model;
- On the computer-use benchmark OSWorld 2.0, it outperforms every other model at any given cost, surpassing Fable 5's best result at just over a third of the cost.
An Agent That Verifies Its Own Work
Opus 5 is markedly stronger at verifying its work and iterating carefully until it succeeds. Anthropic shared several cases: in one Frontier-Bench task where Opus 5 was asked to rebuild a machine part drawing into a 3D FreeCAD model but was deliberately given no way to directly view the drawing, it responded by writing its own computer vision pipeline to extract geometry from raw pixels and successfully reconstructed the part — no competing model with the same setup could solve it after five attempts.
This autonomy is valuable for long-horizon work. Zapier CEO Wade Foster noted that after receiving a raw account-health workbook, Opus 5 ran the full churn-prevention sequence end to end: flagging at-risk accounts, alerting the right owner, and summarizing for retention ops. Previous models didn't pass; Opus 5 hit 100%. For automated workflows, this means fewer failed retries and less manual intervention.
Alignment and Safety Strategy
Anthropic says Opus 5 is "our most aligned model to date": it adheres to Claude's Constitution better than Opus 4.8, Sonnet 5, or Fable 5, exhibits the lowest rates of deceptive behavior, is the least susceptible to being tricked into misuse, and is its safest model yet in avoiding reckless actions with hard-to-reverse side effects. On dual-use biology and cybersecurity capability evaluations, Opus 5 does not advance the frontier in risky capabilities and remains behind Mythos 5.
On safeguards, Opus 5's cyber classifiers are proportionally about 85% less restrictive than Fable 5's — they allow finding vulnerabilities in source code but block binary-based vulnerability scanning, penetration testing, and exploit generation; flagged requests fall back to Opus 4.8 by default. Combined with the new API automatic fallbacks and mid-conversation tool changes (which do not invalidate the prompt cache), enterprises get better availability while staying safe.
Availability and Integration
Opus 5 is available on all platforms today, becoming the default model on Claude Max and the strongest model on Claude Pro, with the API model name claude-opus-5. It also offers Fast mode, running at about 2.5x the default speed for twice the base price, suitable for latency-sensitive scenarios. For developers, Anthropic also released two beta features: mid-conversation tool changes (switch available tools without invalidating the prompt cache) and API automatic fallbacks (requests flagged by safety classifiers automatically route to another model).
If you are considering integrating Opus 5 into your business, start by combining model fine-tuning and model deployment practices, run an evaluation on your real data, and then decide which tier of your workflows it belongs in.
Pricing and tiers at a glance
Putting the numbers side by side makes it clearer:
| Model | Input (per million tokens) | Output (per million tokens) | Notes |
|---|---|---|---|
| Opus 4.8 | $5 | $25 | Previous flagship |
| Opus 5 | $5 | $25 | Same price, far stronger |
| Opus 5 Fast | $10 | $50 | ~2.5x faster |
| Fable 5 | $10 | $50 | Higher flagship tier |
For latency-sensitive interactions (streaming answers in a support chat, for example), paying double the token cost for roughly 2.5x speed is usually worth it; for batch jobs, stick with the default tier.
A minimal integration example
Developers can call the model directly by its API name claude-opus-5:
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{
"model": "claude-opus-5",
"max_tokens": 1024,
"messages": [{"role": "user", "content": "Draft an FAQ structure for this page"}]
}'
Before rolling it into production, run a side-by-side evaluation of Opus 5 and your current model on the same set of real tasks rather than trusting benchmark scores alone — see model evaluation for methods.
Reference: Anthropic API docs https://docs.anthropic.com/en/api/messages
16IDC Take
Claude Opus 5 sends a signal similar to GPT-5.6: leading model vendors are shifting from piling on capability to squeezing cost. Opus 5 delivers near-flagship capability at last generation's price, effectively lowering the barrier to "frontier intelligence" by another notch. For smaller teams and independent sites, this means getting near-top-tier coding and agent capability at lower API cost — with significantly more choice.