For weeks, the coding conversation was all Fable 5.
Expensive. Guardrailed. Worth it for the hardest agent runs, maybe. But your wallet felt every multi-file refactor.
Then on July 24, 2026, Anthropic dropped Claude Opus 5 and quietly rewrote the default answer.
Opus 5 scores within a hair of Fable 5 on real coding benchmarks like CursorBench 3.2, beats it outright on Frontier-Bench v0.1, and costs half as much per task. In Cursor, it even works with Zero Data Retention, no Fable 5 privacy opt-in required.
So the real question is not "Is Opus 5 good?" It is: should Opus 5 replace Fable 5 as your daily coding model?
This guide covers everything developers need to know about Claude Opus 5 for coding: what changed, where it wins, how to enable it in Cursor and Claude Code, and when you still reach for Fable 5 anyway.
Quick answer: Claude Opus 5 is Anthropic's best everyday coding model as of July 2026. It matches Fable 5 on CursorBench 3.2 (66.7% vs 66.5% at default effort in Cursor) at roughly half the cost, doubles Opus 4.8 on Frontier-Bench v0.1, and ships with thinking on by default plus a 1M token context window. Use API id
claude-opus-5in Claude Code, pick it from Cursor's model menu, or enable it in GitHub Copilot. Pricing: $5/M input, $25/M output (same as Opus 4.8). For hard autonomous runs only, Fable 5 still has an edge. For most coding, Opus 5 is the smarter default.
In this guide
- What is Claude Opus 5 for coding?
- July 24 launch: why it matters now
- Opus 5 vs Fable 5 vs Opus 4.8
- Benchmarks that actually map to real work
- How to use Opus 5 in Cursor
- How to use Opus 5 in Claude Code
- Opus 5 in GitHub Copilot
- Pricing, effort, and Fast mode
- When to use Opus 5 vs other models
- Frequently asked questions
What is Claude Opus 5 for coding?
Claude Opus 5 is Anthropic's flagship Opus-class model, released July 24, 2026. Anthropic positions it as near-Fable 5 intelligence at Opus speed and Opus pricing.
For developers, that translates to a model built for:
- Agentic coding: multi-step tool loops, multi-file edits, regression checks
- Long-horizon tasks: holding context across extended sessions without losing the thread
- Root-cause debugging: tracing bugs across a codebase instead of patching symptoms
- Everyday shipping: feature work, refactors, test writing, PR-ready output
Technical specs that matter for coding:
| Spec | Claude Opus 5 |
|---|---|
| API model ID | claude-opus-5 |
| Context window | 1 million tokens |
| Knowledge cutoff | May 2026 |
| Thinking | On by default (adaptive) |
| Effort control | low, medium, high, xhigh, max |
| Input pricing | $5 per million tokens |
| Output pricing | $25 per million tokens |
| Fast mode | ~2.5x speed at 2x price ($10/$50 per M tokens) |
Unlike Fable 5, Opus 5 does not require a separate data retention opt-in in Cursor. That alone makes it the easier pick for teams with strict privacy settings.
July 24, 2026: why Opus 5 changed the coding stack
Three things landed the same week and collided:
- Fable 5's promo window ended (July 19). Pro users shifted to usage credits for Fable 5. See our Fable 5 $100 credits guide.
- Opus 5 launched (July 24) as the new default on Claude Max and the strongest model on Claude Pro.
- Cursor added Opus 5 same day, calling it a Fable 5 match at half the price with Zero Data Retention compatibility.
The pre-launch consensus on X was that Opus 5 would land "comparable to Fable 5, but won't surpass it." The coding benchmarks told a different story. Opus 5 cleared Fable 5 on Frontier-Bench v0.1 by a wide margin while costing half as much per task on CursorBench-style work.
If you read our best new AI tools for founders (July 2026) roundup, Opus 5 is the update that reshuffles the whole stack.
Claude Opus 5 vs Fable 5 vs Opus 4.8 for coding
Here is the honest three-way comparison developers actually need.
| Factor | Opus 5 | Fable 5 | Opus 4.8 |
|---|---|---|---|
| CursorBench 3.2 (peak) | Within 0.5% of Fable 5 at max effort | Top tier on hardest agent runs | Noticeably behind |
| Frontier-Bench v0.1 | 43.3% (leads) | 33.7% | 18.7% |
| Cost per task | ~Half of Fable 5 | Highest ($10/$50 per M tokens) | Same list price as Opus 5 |
| Thinking default | On by default | Thinking variant available | Adaptive thinking optional |
| Cursor privacy | Zero Data Retention compatible | Requires data retention opt-in | Standard |
| Guardrail fallbacks | ~85% fewer interventions vs Fable 5 | More frequent safety routing | Baseline |
| Best for | Daily coding, agent loops, debugging | Long autonomous runs, max-hard problems | Legacy fallback |
Plain English: Opus 4.8 is the previous generation. Fable 5 is the specialist for the nastiest agentic work. Opus 5 is the model you leave selected 80% of the time.
Anthropic still positions Claude Fable 5 ahead on certain high-risk specialized tasks (cybersecurity, some long-horizon research). For standard software engineering, Opus 5 is the new state of the art.
Claude Opus 5 coding benchmarks explained
Benchmarks lie sometimes. These two map closest to what paid developers actually do.
Frontier-Bench v0.1 (agentic terminal coding)
Real software engineering tasks: multi-file changes, debugging, building features from specs. Not LeetCode trivia.
| Model | Score |
|---|---|
| Claude Opus 5 | 43.3% |
| Claude Fable 5 | 33.7% |
| Claude Opus 4.8 | 18.7% |
Opus 5 more than doubles Opus 4.8 and beats Fable 5 by nearly 10 points. That is the headline number for "Claude Opus 5 for coding."
CursorBench 3.2 (real-world IDE work)
Cursor's benchmark measures coding inside the environments developers actually use.
At default effort in Cursor:
- Opus 5: 66.7%
- Fable 5: 66.5%
Essentially tied. At max effort, Anthropic reports Opus 5 within 0.5% of Fable 5's peak while costing half as much per task.
Scott Wu at Cognition (Devin) confirmed early testing showed Opus 5 approaching Fable-level performance at half the cost, with particular strength in debugging and root-cause analysis.
Takeaway: Run your own eval on your repo for final decisions. Benchmarks tell you direction. Your codebase tells you truth.
How to use Claude Opus 5 in Cursor
Cursor announced Opus 5 availability on July 24, 2026.
Step 1: Update Cursor
Model releases ship in app updates. Check Help > Check for Updates or download from cursor.com.
Step 2: Select Opus 5 in the model picker
- Open Chat or Agent mode
- Open the model dropdown
- Choose Claude Opus 5
No Fable 5 data retention approval needed. Opus 5 is compatible with Zero Data Retention policies that block Fable 5 by default.
Step 3: Match effort to the task
- Routine edits, boilerplate, small fixes: default or medium effort
- Multi-file refactors, tricky bugs: high or xhigh effort
- Maximum agent autonomy: max effort (watch token spend)
Step 4: Switch to Fable 5 only when stuck
If Opus 5 hits a guardrail or stalls on an unusually complex autonomous migration, switch to Fable 5 for that session. Cursor may also auto-fallback in some Fable 5 guardrail cases. Opus 5 should be your default, not your only tool.
Building a product with this stack? Our Micro SaaS launch playbook pairs well with Cursor + Opus 5 for solo founder MVPs.
How to use Claude Opus 5 in Claude Code
Claude Code got Opus 5 on launch day. Max subscribers receive it as the default. Pro users get it as the strongest model on their tier.
Quick model switch (current session)
Run in an active Claude Code session:
/model
Pick Claude Opus 5 from the list.
Or pass the ID directly:
/model claude-opus-5
Launch flag (one session)
claude --model claude-opus-5
Set as permanent default
Add to your shell config (zsh example):
export ANTHROPIC_MODEL="claude-opus-5"
See Claude Code model configuration for bash and Windows setup.
Effort and thinking tips
On Opus 5, thinking is on by default. The effort parameter controls depth:
| Effort | Use for |
|---|---|
| medium | Most daily coding |
| high | Hard debugging, architecture decisions |
| xhigh / max | Long agent runs, complex refactors |
Breaking change vs Opus 4.8: You cannot disable thinking at xhigh or max effort. Setting thinking: {"type": "disabled"} with those effort levels returns a 400 error. See Anthropic's Opus 5 docs.
Fast mode for tight loops
Fast mode runs ~2.5x faster at 2x price ($10/$50 per million tokens). Useful for interactive edit-run-fix cycles. Skip it for overnight agent jobs where latency does not matter.
Requires Claude Code 2.1.219 or later.
Claude Opus 5 in GitHub Copilot
GitHub added Opus 5 on July 24, 2026 for Copilot Pro+, Business, and Enterprise users.
Available in:
- VS Code and Visual Studio
- Copilot CLI and cloud agent
- JetBrains, Xcode, Eclipse
- github.com and GitHub Mobile
Early testing showed strong performance on autonomous code changes, regression verification, and multi-tool coordination. Admins on Business/Enterprise plans must enable the Opus 5 policy in Copilot settings.
Billed at provider API list price under usage-based billing. See GitHub Copilot pricing for models.
Claude Opus 5 pricing, effort, and Fast mode
Standard pricing (same as Opus 4.8)
| Input | Output | |
|---|---|---|
| Standard | $5 / million tokens | $25 / million tokens |
| Fast mode | $10 / million tokens | $50 / million tokens |
Fable 5 costs $10/$50 per million tokens. Opus 5 at standard speed matches half the output cost of Fable 5.
Cursor billing
On Cursor Pro, Opus 5 draws from your plan's model pool at standard Opus rates (not Fable 5's premium API pool). That makes it significantly cheaper for daily agent use than defaulting to Fable 5.
If you still have Fable 5 credits from July's transition, see how to claim Fable 5 $100 credits. Many developers will burn those faster on Fable 5 than they would on Opus 5 for the same tasks.
Cost control tips
- Use medium effort for 80% of tasks
- Reserve max effort for agent runs that justify the token bill
- Enable Fast mode only when waiting on the model blocks your loop
- Prompt caching and batch API discounts apply on the Claude Platform
When to use Opus 5 vs Fable 5 vs Sonnet for coding
Not every task needs the most expensive brain in the room.
| Task | Best model |
|---|---|
| Daily feature work, bug fixes, PRs | Opus 5 |
| Root-cause debugging across a large repo | Opus 5 (high/xhigh effort) |
| Long autonomous migration or multi-hour agent job | Fable 5 (max effort) |
| Quick autocomplete, small edits, docs typos | Sonnet 5 or lighter model |
| Strict Zero Data Retention in Cursor | Opus 5 |
| Budget-sensitive Pro plan usage | Opus 5 over Fable 5 |
Decision rule: Start on Opus 5. Escalate to Fable 5 only when Opus 5 stalls or the task is genuinely multi-hour autonomous work. Drop to Sonnet for speed and cost on trivial edits.
For broader model strategy beyond coding, read ChatGPT vs Claude in 2026.
Frequently asked questions
Is Claude Opus 5 good for coding?
Yes. Claude Opus 5 is Anthropic's strongest everyday coding model as of July 2026. It leads Frontier-Bench v0.1 (43.3%), matches Fable 5 on CursorBench 3.2 within 0.5% at max effort, and more than doubles Opus 4.8's coding scores at the same list price.
Is Opus 5 better than Fable 5 for coding?
For most software engineering tasks, yes. Opus 5 beats Fable 5 on Frontier-Bench v0.1 and ties on CursorBench 3.2 at roughly half the cost per task. Fable 5 still leads on certain specialized high-risk domains and some longest-horizon autonomous research workloads.
How do I use Claude Opus 5 in Cursor?
Update Cursor, open the model picker in Chat or Agent mode, and select Claude Opus 5. No Fable 5 data retention opt-in is required. Cursor confirmed Opus 5 is compatible with Zero Data Retention policies.
What is the Claude Opus 5 API model ID?
Use claude-opus-5 on the Claude API, in Claude Code (/model claude-opus-5), and in configuration files. On Amazon Bedrock the ID is anthropic.claude-opus-5.
How much does Claude Opus 5 cost for coding?
Standard pricing is $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.8 and half of Fable 5's output rate. Fast mode costs $10/$50 per million tokens for approximately 2.5x faster responses.
Does Opus 5 have thinking enabled by default?
Yes. Claude Opus 5 runs with adaptive thinking on by default. Control depth with the effort parameter (low through max). Disabling thinking is only allowed at high effort or below, a breaking change from Opus 4.8.
Is Claude Opus 5 available in Claude Code?
Yes, from July 24, 2026. Max subscribers get it as the default. Pro subscribers access it as the top model on their tier. Select it with /model or claude --model claude-opus-5.
Is Claude Opus 5 available in GitHub Copilot?
Yes. GitHub rolled out Opus 5 on July 24, 2026 for Copilot Pro+, Business, and Enterprise users across VS Code, Copilot CLI, JetBrains, and other supported clients. Rollout is gradual.
When should I still use Fable 5 instead of Opus 5?
Use Fable 5 for the hardest multi-hour agent runs, tasks where Opus 5 stalls, or specialized workloads where Anthropic still positions Fable 5 ahead. For daily coding, debugging, and feature development, Opus 5 is the better default.
Opus 5 vs Opus 4.8 for coding: is the upgrade worth it?
Yes, if you do agentic or multi-file work. Opus 5 more than doubles Opus 4.8 on Frontier-Bench v0.1 at the same price. The upgrade is free at the API level. You pay in tokens, not a higher per-token rate.
Bottom line
Claude Opus 5 is the first Opus model that feels like a credible daily driver for coding, not just a fallback when Fable 5 is too expensive or too locked down.
It matches Fable 5 where most developers live (IDE agent work), beats it on agentic terminal benchmarks, costs half as much per task, and slides into Cursor without the Fable 5 privacy friction.
Switch your default today. Keep Fable 5 in the garage for the jobs that actually need it.
Last updated: July 27, 2026. Sources: Anthropic Opus 5 announcement, Cursor forum, GitHub Changelog, Anthropic Opus 5 docs. Corrections? Contact us. More guides on our playbooks index or newsletter.
Related reads: How to Claim Fable 5 Credits · Best New AI Tools for Founders (July 2026) · ChatGPT vs Claude 2026 · Micro SaaS Launch Playbook
More playbooks
PUBLISHED JUL 25, 2026 | 16 MIN
Best New AI Tools for Founders (July 2026)
The best new AI tools for founders in July 2026, from OpenWorker and Claude Fable 5 to Cursor, Perplexity, and Trooly. Honest picks by use case, pricing, and what actually saves time when you're building.
PUBLISHED JUL 22, 2026 | 17 MIN
Micro SaaS Launch Playbook: Zero to First 10 Customers (2026)
A practical micro SaaS launch playbook to go from zero to your first 10 paying customers. Niche selection, MVP scope, pricing, distribution, outreach scripts, and a 30-day plan. No vanity metrics.
PUBLISHED JUL 21, 2026 | 14 MIN
How to Claim Fable 5 $100 Credits (Step-by-Step 2026)
Learn how to claim your free $100 Fable 5 usage credits in Cursor or Claude Pro. Step-by-step guide, eligibility, claim deadline, expiry date, and what happens when credits run out.