Claude Sonnet 5 vs Opus 4.8: Price, Limits, and Benchmarks¶
For / Key Points
For: Engineers, development teams, and Claude Code users deciding when to use Sonnet versus Opus
Key Points:
- Sonnet 5 launches at 2/10 and then moves to 3/15, below Opus 4.8's 5/25
- Benchmarks are close to Opus 4.8, but Opus still leads on several broad agentic evaluations
- Cybersecurity work remains a clear exception: Anthropic recommends Opus 4.8 when reduced guardrails are needed
On June 30, 2026, Anthropic released Claude Sonnet 51. The introductory API price is $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it moves to 3/151. The practical question is simple: which work should move to Sonnet 5, and where should Opus 4.8 remain reserved?
The answer is not a clean replacement. Use Sonnet 5 as the default for everyday coding, research, and Claude Code execution. Keep Opus 4.8 for deep design decisions, hard debugging, and some legitimate cybersecurity workflows.
Bottom Line¶
Sonnet 5 is best understood as near-Opus capability at Sonnet-tier pricing. Anthropic describes it as closing the gap with Opus 4.8 while remaining cheaper, and it is now the default model for Free and Pro plans1.
Usage-limit visibility also matters. Claude Help Center says Pro, Max, Team, and seat-based Enterprise users can view five-hour session usage and weekly usage limits in Settings > Usage, including weekly reset information for "Opus only" and "all other models"2.
That makes Sonnet 5 useful for more than cost reduction. It lets teams keep day-to-day work on the cheaper default path while reserving Opus for places where deeper reasoning is worth the spend and quota pressure.
Pricing¶
At standard pricing, Sonnet 5 costs about 60% of Opus 4.8. The introductory price runs through August 31, 2026, and the standard price remains below Opus 4.81.
| Item (per million tokens) | Sonnet 5 (intro through Aug. 31, 2026) | Sonnet 5 (standard) | Opus 4.8 |
|---|---|---|---|
| Input | $2 | $3 | $5 |
| Output | $10 | $15 | $25 |
The price table is not the whole cost story. Anthropic says Sonnet 5 uses an updated tokenizer, so the same input may map to roughly 1.0-1.35x as many tokens depending on content type1.
The introductory price is designed to make the transition roughly cost-neutral during launch. Teams planning production use after September 1, 2026 should measure real token growth on their own prompts, not just compare list prices.
Benchmarks¶
The official benchmark table shows a major jump from Sonnet 4.6, but not a full Opus 4.8 replacement. Opus 4.8 still leads on SWE-bench Pro, Terminal-Bench 2.1, and OSWorld-Verified1.
| Benchmark | Sonnet 4.6 | Sonnet 5 | Opus 4.8 |
|---|---|---|---|
| SWE-bench Pro | 58.1% | 63.2% | 69.2% |
| Terminal-Bench 2.1 | 67.0% | 80.4% | 82.7% |
| Humanity's Last Exam (no tools) | 34.6% | 43.2% | 49.8% |
| Humanity's Last Exam (with tools) | 46.8% | 57.4% | 57.9% |
| OSWorld-Verified | 78.5% | 81.2% | 83.4% |
| GDPval-AA v2 | 1395 | 1618 | 1615 |
The interesting exception is GDPval-AA v2. Sonnet 5 scores 1618, slightly above Opus 4.8 at 16151. For knowledge-work tasks, the cheaper model may deliver more than enough capability.
The better reading is not "which model wins the table." Anthropic says effort settings let Sonnet 5 move along a cost-performance curve, and high-effort Sonnet 5 can match Opus 4.8 capability levels on some tasks1.
Cybersecurity¶
Cybersecurity is the clearest area where Sonnet 5 should not be treated as equivalent to Opus 4.8. Anthropic says Sonnet 5 was not deliberately trained for cybersecurity tasks and recommends Opus 4.8 for cybersecurity work that requires reduced guardrails1.
The Firefox 147 exploit-development evaluation makes the gap visible14.
| Firefox 147 exploit-development evaluation | Sonnet 4.6 | Sonnet 5 | Opus 4.8 |
|---|---|---|---|
| Working exploit | 0.0% | 0.0% | 8.8% |
| Register control only | 8.8% | 13.2% | 68.8% |
Sonnet 5 did not produce a full working exploit in this evaluation, and its partial-success rate remains far below Opus 4.8. That reflects both safety posture and training target.
This does not mean Sonnet 5 has weaker protections. Anthropic says Sonnet 5 ships with the same real-time cyber safeguards used in Opus 4.7 and Opus 4.81.
Claude Code Use¶
In Claude Code, the practical pattern is Sonnet by default and Opus for high-value reasoning turns. Claude Help Center describes Sonnet as the right choice for the large majority of coding work, while Opus is better suited to broad refactors, hard debugging, and architecture decisions3.
A useful operating model is:
- Sonnet 5: features, tests, known bugs, routine refactors
- High-effort Sonnet 5: research, longer implementations, multi-file changes
- Opus 4.8: broad design decisions, difficult defects, legitimate security review
/model opusplan: plan with Opus, execute with Sonnet
Leaving Opus on by default is expensive. The Help Center says Opus costs several times more per turn than Sonnet and can drain limits quickly when used for routine work3.
Summary¶
Sonnet 5 is not simply a cheaper Opus 4.8. It is the model that makes it practical to reserve Opus 4.8 for the work that actually needs it.
For most coding and agentic workflows, Sonnet 5 should become the default. For difficult reasoning, hard debugging, and legitimate cybersecurity work, Opus 4.8 still has a clear role.
The model-selection question is shifting. It is no longer only "which model is smartest?" It is where should the expensive reasoning budget be preserved? Sonnet 5 makes that budget allocation more practical.
Related Articles¶
- Claude Sonnet 4.6 Release - Baseline for the previous Sonnet generation
- Claude Opus 4.6 Complete Guide - Context for evaluating Opus-class models
- Claude Fable 5 Launch - A look at the capability tier above Opus