2026-06-30
Claude Opus 4.8 vs Sonnet 5 — Which Should You Actually Use?
Anthropic shipped Claude Sonnet 5 on June 30, 2026 — and the headline everyone latched onto is that it runs close to the flagship Claude Opus 4.8 for a fraction of the cost. So which one should you point your work at? Here's the honest breakdown, including what developers are actually saying about it on X (Twitter).
The short version
- Sonnet 5 is the new default for almost everyone — near-Opus quality on coding and agentic work, at Sonnet prices. It's now the default model in Claude Code for Pro users.
- Opus 4.8 is still the ceiling. When a task is genuinely hard — long-horizon agents, the trickiest debugging, the highest-stakes reasoning — Opus at high effort wins.
- For most people, most of the time, start on Sonnet 5 and reach for Opus 4.8 when you hit its limits.
What X is saying
Anthropic's own accounts set the framing on launch day:
"Sonnet 5 is a substantial improvement over Sonnet 4.6 on reasoning, tool use, coding, and knowledge work. Its performance is close to Opus 4.8, at lower prices." — @claudeai
"Claude Sonnet 5 is here. Top-tier performance on coding and tool use at Sonnet pricing, with a 1M context window. It's the new default in Claude Code for Pro users, and available everywhere on the Claude Platform, including the API and Managed Agents." — @ClaudeDevs
The tool builders who plugged it in the same day were largely convinced. Cursor called Sonnet 5 a "meaningful step up," posting a CursorBench jump to 57% from Sonnet 4.6's 49%. Cognition (the team behind Devin) said Sonnet 5 actually outperforms Opus 4.8 on their FrontierCode Extended benchmark. And Cline highlighted Opus 4.8-level results on Terminal-Bench "for less than half the cost."
But it wasn't universal applause — which is exactly why the price/performance story deserves a second look. Developer Theo Browne pointed out that on the Artificial Analysis Intelligence Index, Sonnet 5 sometimes cost more than Opus 4.8 to run a full evaluation, because it "thinks" harder (more tokens) to reach the same answer — so the cheaper sticker price doesn't always mean a cheaper job. Others, like @kimmonismus, argued the gains look more like a "Sonnet 4.8/4.9" step than a true "5.0."
Both takes are fair. Sonnet 5 is a real jump — and the version number is doing some marketing work.
Price: the headline difference
This is where the gap is widest, and it's the whole reason Sonnet 5 exists.
| Model | Input / 1M tokens | Output / 1M tokens | Context | Max output |
|---|---|---|---|---|
| Claude Opus 4.8 | $5.00 | $25.00 | 1M | 128K |
| Claude Sonnet 5 | $3.00 | $15.00 | 1M | 128K |
And there's an introductory discount: through August 31, 2026, Sonnet 5 runs at $2 / $10 per million tokens. That's roughly 60% cheaper than Opus 4.8 on the sticker — though, as Theo's point above shows, watch your actual token usage on reasoning-heavy jobs before assuming the bill drops by the same margin.
Benchmarks: how close is "close"?
Anthropic's launch numbers tell a consistent story — Opus 4.8 leads on the hardest agentic coding, but Sonnet 5 is neck-and-neck (and occasionally ahead) elsewhere:
| Benchmark | Sonnet 5 | Opus 4.8 |
|---|---|---|
| SWE-bench Pro (agentic coding) | 63.2% | 69.2% |
| GDPval-AA v2 (knowledge work) | 1,618 | 1,615 |
| OSWorld-Verified (computer use) | 81.2% | 78.5% |
| Humanity's Last Exam (with tools) | 57.4% | 57.9% |
The ~6-point SWE-bench Pro gap is the clearest signal that Opus 4.8 is still the strongest pure coding agent. But on knowledge work and computer use, Sonnet 5 edges ahead — which is remarkable for a model at half the price.
Where each one pulls ahead
Reach for Opus 4.8 when:
- The task is a long-horizon, mostly-autonomous run (big refactors, overnight coding jobs).
- You need the highest ceiling on the hardest reasoning or debugging.
- Correctness matters more than cost, and you can run it at high or
xhigheffort.
Reach for Sonnet 5 when:
- You're doing everyday coding, tool use, and agent work — most of what most people do.
- You're running lots of calls and cost adds up (support bots, pipelines, batch jobs).
- You want strong computer-use and knowledge-work performance without the flagship price.
Both share the same modern foundation: a 1M-token context window, 128K max output, adaptive thinking, and the same tool and agent surface on the Claude Platform.
Which should you use?
For nearly everyone, Sonnet 5 is the right default — it's what Anthropic ships in Claude Code for Pro users for a reason. Keep Opus 4.8 in your back pocket for the jobs that genuinely need the extra headroom.
The best way to feel the difference is to run both through a great tool. If you write code, start with our best AI coding assistants ranking — Cursor and Claude Code both run Sonnet 5 brilliantly and let you switch up to Opus 4.8 in a click. For everything else, see the guides to the best tools for writing, building apps, and research with Claude.
Reaction quotes above are from public posts on X and launch-day coverage; benchmark figures are from Anthropic's Claude Sonnet 5 and Opus 4.8 announcements. Confirm the latest pricing on Anthropic's pricing page.