Claude Sonnet 5 Is Here: Pricing, 1M Context, and What Changes for Developers

On June 30, 2026, Anthropic shipped Claude Sonnet 5 — and it landed everywhere at once: Claude.ai, the Claude Developer Platform, Claude Code, and every subscription tier from Free to Max. If you've been putting off a model upgrade because switching felt risky or expensive, this release removes both excuses. Sonnet 5 is cheaper than its predecessor, faster on agentic tasks, and it's now the default model in Claude Code — meaning most developers are already running it whether they opened a changelog or not.

This matters more than a routine point release. Anthropic is explicitly positioning Sonnet 5 as "the cheaper way to run agents" — a direct signal that long-running, tool-heavy, multi-step agent workflows are the primary design target, not just chat quality benchmarks. If you're building anything with Claude Code, the Claude Agent SDK, or MCP servers, the economics of your architecture just shifted.

Here's exactly what changed, what it costs, and how to decide whether — and how — to move your workloads over.

What's Actually New in Claude Sonnet 5

Three changes matter most for anyone building with the API or using Claude Code day to day:

Anthropic describes Sonnet 5 as its "most agentic Sonnet model yet," with specific gains in tool use, autonomous multi-step task handling, and coding benchmarks. Performance is being pitched as close to Opus 4.8 for a fraction of the cost — the same value proposition that made Sonnet 4.6 the default choice for teams that needed near-frontier quality without frontier pricing.

Claude Sonnet 5 vs. Opus 4.8: Which One Should You Actually Use?

The question every team asks after a Sonnet release: do I still need Opus? The honest answer depends on what your workload looks like.

If your work is agentic — running Claude Code sessions, orchestrating subagents, executing tool calls in loops, processing large batches of similar tasks — Sonnet 5 is very likely the better default now. Reserve Opus 4.8 for the minority of tasks that genuinely need deeper multi-step reasoning under ambiguity: complex architectural decisions, adversarial code review, or judgment calls where a wrong answer is costly. This mirrors the guidance in our Claude model selection guide — route by task difficulty, not habit.

How to Start Using Claude Sonnet 5 Today

You don't need to do anything to get Sonnet 5 in Claude Code — it's the new default as of this release. But if you're calling the API directly or want to confirm your setup, here's the practical checklist:

1. Check your Claude Code version and model

bashclaude --version

If you're on a recent build, Sonnet 5 is already active. Confirm inside a session:

/model

This shows the currently active model and lets you switch to Opus 4.8 for a single session if a task demands deeper reasoning.

2. Update API calls to the new model ID

If you're calling the Claude API directly (not through Claude Code), point requests at the Sonnet 5 model identifier:

pythonimport anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=4096,
    messages=[
        {"role": "user", "content": "Refactor this module for readability."}
    ]
)

Swap in your existing prompt and tool definitions — Sonnet 5 supports the same tool-use schema as Sonnet 4.6, so migrations for existing agents and MCP integrations should be close to drop-in.

3. Re-benchmark your prompt caching strategy

With a 1M-token context window at standard pricing, some teams that built elaborate retrieval-augmented pipelines specifically to work around smaller context limits may find those pipelines are now unnecessary overhead. Before you rip anything out, benchmark: for latency-sensitive, high-frequency queries, targeted retrieval is still often cheaper and faster than stuffing a full corpus into context on every call. Use the large window for depth (one-time large-document analysis, big refactors) and keep retrieval for high-throughput repeated queries.

4. Watch your token spend during the promo window

$2/$10 pricing is temporary. If you're planning a large one-time job — bulk document processing, a full codebase audit, generating a large test suite — running it before August 31, 2026 costs meaningfully less than waiting. Budget accordingly if your project timeline has flexibility.

What This Means If You're Studying for the Claude Certified Architect (CCA) Exam

If you're prepping for CCA-F certification, model releases like this aren't just news — they're exam-relevant. Certification content increasingly tests whether you understand model selection reasoning, not just API syntax. Expect scenario questions along the lines of: "A team needs to process a 500-page technical document and generate a structured migration plan — which model and configuration minimizes cost while meeting the context requirement?" Sonnet 5's 1M window at standard pricing is exactly the kind of platform detail that changes the "textbook correct" answer to these questions.

Two things worth doing now if you're on a CCA study track:

• Update your mental pricing model. If you memorized Sonnet 4.6's cost structure, note that Sonnet 5 currently mirrors it during the promotional window — but the underlying capability (context window, agentic tool use) has moved up a tier. Certification questions test current platform capabilities, not last quarter's.
• Practice model-routing scenarios. Given a workload description, can you justify Sonnet 5 vs. Opus 4.8 vs. Haiku in one sentence citing cost, latency, and reasoning depth? This is a recurring question pattern in architect-level assessments.

Other Anthropic News from the Same Week

Sonnet 5 wasn't released in isolation. In the same week, Anthropic also:

• Lifted export controls on Claude Fable 5 and Mythos 5 after a standoff with the Trump administration's Commerce Department, making those models available globally via the Claude Platform, Claude.ai, Claude Code, and Claude Cowork starting July 1, 2026.
• Launched Claude Science, a new AI workbench product for biopharma and scientific research, alongside Anthropic's own in-house drug discovery programs — a clear signal of expansion beyond general-purpose coding and chat use cases ahead of a possible IPO.
• Partnered with California under Governor Newsom to bring Anthropic tools to state agencies, continuing Anthropic's push into public-sector deployments.

Taken together, these moves show Anthropic diversifying both its model lineup and its go-to-market surface area — general agents (Sonnet 5), specialized science tooling (Claude Science), and government partnerships, all in a single week.

Key Takeaways

• Claude Sonnet 5 launched June 30, 2026, with $2/$10 per MTok intro pricing through August 31, 2026, a 1M-token context window at standard pricing, and 128k max output tokens.
• It's now the default model in Claude Code and across Free, Pro, and Max plans — no manual switch required.
• For agentic, tool-heavy workloads, Sonnet 5 is likely a better default than Opus 4.8; reserve Opus for high-ambiguity reasoning tasks.
• Existing Sonnet 4.6 API integrations should migrate with minimal changes — same tool-use schema, updated model ID.
• If you're studying for the Claude Certified Architect exam, expect model-selection scenario questions to reflect Sonnet 5's new capability tier.

Next Steps

Model-routing questions like "Sonnet 5 or Opus 4.8?" show up constantly on the Claude Certified Architect (CCA-F) exam. If you want structured practice with real scenario questions instead of guessing from changelogs, check out AI for Anything's CCA practice test bank — it's built specifically around the kind of platform-knowledge questions Anthropic's certification actually asks, updated as new models like Sonnet 5 ship.

• Anthropic Newsroom
• TechCrunch: Anthropic launches Claude Sonnet 5 as a cheaper way to run agents
• Releasebot: Claude Code Updates by Anthropic - July 2026
• CNBC: Anthropic says Trump admin has lifted export controls on Claude Fable 5 and Mythos 5
• Endpoints News: Anthropic launches Claude Science