Claude Sonnet 5 Released: The Most Agentic Sonnet Yet, Near Opus 4.8 at a Lower Price

On June 30, 2026, Anthropic released Claude Sonnet 5 (internal codename Fennec), calling it "the most agentic Sonnet model yet." It can plan, use tools like browsers and terminals, and run autonomously at a level that, just months ago, required larger and more expensive models. The headline: its overall performance now approaches flagship Opus 4.8, while costing significantly less. For anyone who codes, automates, or does knowledge work with AI daily, that's a real jump in value.

What's new: a mid-tier model that gets work done on its own

We used to judge models on whether they answered correctly. This generation is judged on whether it can finish an entire job by itself — that's the agentic capability Anthropic leaned into. Sonnet 5:

• Plans — given a complex goal (say, "migrate this project from Vue 2 to Vue 3"), it breaks it into ordered steps and works through them, instead of one-shot answering.
• Uses tools — natively handles browsers, terminals, and file systems: it can research, run commands, read/write files, and run tests across a full workflow.
• Runs autonomously — stays focused across dozens or hundreds of steps without drifting, cutting the cost of babysitting and constant correction.
• Stronger reasoning, tool use, coding and knowledge work than Sonnet 4.6 — across the board, not a single metric.

Benchmarks: within a hair of Opus 4.8

Per Anthropic's published comparison (official figures take precedence), Sonnet 5 leaps past Sonnet 4.6 and even edges out flagship Opus 4.8 on knowledge work:

Benchmark	Sonnet 4.6	Sonnet 5	Opus 4.8
SWE-bench Pro (agentic coding)	58.1%	63.2%	69.2%
GDPval-AA v2 (knowledge work)	1,395	1,618	1,615
Humanity's Last Exam (with tools)	—	57.4%	57.9%

It narrows the agentic-coding gap to six points, slightly surpasses Opus 4.8 on GDPval knowledge work, and essentially ties Opus on tool-assisted reasoning (57.4% vs 57.9%). For most everyday and production tasks, Sonnet 5 has moved from "good enough" to "genuinely good," with only the hardest jobs needing Opus.

Pricing: $2 / $10 intro — agent costs roughly halved

Pricing is aggressive. On the Claude Platform (API):

• Through August 31, 2026: $2 per million input tokens, $10 per million output tokens.
• After that: $3 input, $15 output.

Even at standard pricing it's about 60% cheaper than Opus 4.8 while landing in a heavily overlapping tier. Why does that matter for agents? Agentic tasks call tools repeatedly across many turns, burning several to dozens of times the tokens of a normal chat — halving the per-token price changes the whole long-run bill. The API model ID is claude-sonnet-5. Onboard before the late-August increase to lock in the lower rate.

Availability and safety

Sonnet 5 is available today across all plans and is the default for Free and Pro, plus Max, Team and Enterprise. Anthropic reports a lower rate of undesirable behaviors than Sonnet 4.6, a much weaker ability to perform cybersecurity tasks than current Opus models, and cyber safeguards enabled by default — more capability with lower risk shipped together.

Which to choose: Sonnet 5 / Opus 4.8 / Fable 5

• Daily + most production work → Sonnet 5: best value for coding, automation and long-horizon knowledge work; make it your default.
• Hardest coding and reasoning → Opus 4.8: switch up only when Sonnet 5 stalls on a tough problem.
• Absolute frontier → Fable 5: if your task is extremely capability-sensitive, see our Claude Fable 5 deep dive.

In short: default to Sonnet 5, escalate to Opus for hard jobs, look to Fable 5 at the limit.

Get Claude Sonnet 5 on Neuronicx

No need to wrestle with official sign-up, card binding, currency or top-ups — on Neuronicx you buy the product you need and use it instantly, with auto-delivery on payment, available in HK and mainland:

• Claude Pro — the entry point for daily Sonnet 5 use on web and apps.
• Claude Max — higher limits for power users and teams; smoother on long tasks and long context.
• Claude Code — the developer coding subscription where Sonnet 5's agentic coding shines: plan, use the terminal, run tests end to end.
• Claude API and the AI API hub — call claude-sonnet-5 directly in your apps and automations, usage-based and reliable.

Browse all memberships and APIs under AI memberships. We use official channels, deliver instantly, and handle the setup-and-payment hassle so you can focus on the work.

FAQ

How much better is Sonnet 5 than Sonnet 4.6?

Across reasoning, tool use, coding and knowledge work. GDPval jumps from 1,395 to 1,618 and SWE-bench Pro from 58.1% to 63.2% — a sizable step up, especially in autonomous, finish-the-job agentic ability.

Can it replace Opus 4.8?

For most tasks it's close enough and cheaper; for the hardest coding and reasoning, Opus 4.8 still leads. Default to Sonnet 5, switch to Opus for the toughest jobs.

Will the price go up?

Yes. Intro pricing ($2 in / $10 out) runs through August 31, 2026, then moves to $3 / $15. Onboard during the window to lock in savings.

Is Sonnet 5 cost-effective for agents?

Very. Agentic tasks burn far more tokens than chat, and Sonnet 5 delivers near-flagship capability at roughly 40-60% of flagship price — exactly why it's billed as the most agentic Sonnet.

How do I start on Neuronicx?

Pick by use case: Claude Pro/Max for web, Claude Code for development, Claude API or the API hub to build. Instant auto-delivery, available in HK and mainland.