Claude Sonnet 5: What's New, the Pricing, and When to Use It | Aravind Arumugam
ai5 min read
Claude Sonnet 5: What's New, the Pricing, and When to Use It
Anthropic released Claude Sonnet 5 on 30 June 2026, and it is its most agentic Sonnet model yet, built to plan, use tools like browsers and terminals, and run autonomously at a level that until recently needed larger, pricier models. The headline: its performance sits close to the flagship Opus 4.8, but it costs a fraction of the price. For anyone running AI agents at scale, that combination is the whole story.
What is Claude Sonnet 5?
Claude Sonnet 5 (model ID claude-sonnet-5) is the new mid-tier model in Anthropic's line-up, positioned as "the best combination of speed and intelligence". It carries a 1M-token context window, up to 128K tokens of output, and a January 2026 knowledge cutoff. It supports adaptive thinking, and on the API the effort parameter defaults to high.
Anthropic frames it squarely around agents. In its own words, Sonnet 5 "can make plans, use tools like browsers and terminals, and run autonomously at a level that, just a few months ago, required larger and more expensive models".
From 1 July 2026 it becomes the default model for Free and Pro plans
, and it is available to Max, Team, and Enterprise users through the Claude API, Claude Code, Amazon Bedrock, and Google Cloud.
How much does Claude Sonnet 5 cost?
This is where it gets interesting. Sonnet 5 launches with introductory pricing of $2 per million input tokens and $10 per million output tokens, running through 31 August 2026. After that it moves to its standard rate of $3 / $15 per million tokens.
For comparison, Opus 4.8 is priced at $5 / $25 per million tokens. So during the introductory window, Sonnet 5 is roughly 60% cheaper than Opus on both input and output, at a quality level Anthropic says is close behind.
One caveat for cost modelling: Sonnet 5 uses an updated tokenizer that maps the same text to roughly 1.0 to 1.35 times as many tokens as Sonnet 4.6. Re-baseline your token counts before assuming a like-for-like saving.
How does it compare to Opus 4.8 and Sonnet 4.6?
On Anthropic's agentic-coding benchmark, as reported by TechCrunch, the scores line up like this:
Claude Sonnet 5: 63.2%
Claude Opus 4.8: 69.2%
Claude Sonnet 4.6: 58.1%
So Sonnet 5 closes most of the gap to Opus while clearly out-performing the model it replaces. On a knowledge-work benchmark, Anthropic reports Sonnet 5 slightly ahead of Opus 4.8. Across agentic evaluations such as BrowseComp (agentic search) and OSWorld-Verified (computer use), Sonnet 5 sits between Sonnet 4.6 and Opus 4.8 on the cost-performance curve.
The honest read: Opus 4.8 is still the top model for the hardest reasoning. Sonnet 5 is the value pick that gets you most of the way there.
Is Claude Sonnet 5 safer than Sonnet 4.6?
Yes, on most measures. Anthropic reports that Sonnet 5 has lower rates of undesirable behaviour, deception, sycophancy, and hallucination than Sonnet 4.6, and is better at refusing malicious requests and resisting prompt-injection attacks.
Two things to keep in mind. First, on automated misalignment audits, Sonnet 5 is not yet equal to Opus 4.8, which remains the safer model on those metrics. Second, by design it performs substantially worse at developing cyber-exploits, scoring 0.0% on a Firefox 147 exploit-development task. That is a deliberate safety posture, not a gap that affects legitimate work.
When should you use Sonnet 5 instead of Opus 4.8?
A practical rule of thumb:
Reach for Sonnet 5 for high-volume agent loops, tool-heavy workflows, coding agents, and anything where you pay per token across many runs. The cost saving compounds fast.
Stay on Opus 4.8 for the hardest single-shot reasoning, long-horizon autonomous work where a six-point benchmark gap matters, and tasks where the cost of an error outweighs the token saving.
If you are already on Sonnet 4.6, moving to Sonnet 5 is close to a free upgrade in capability. Just re-baseline your token budgets for the new tokenizer first.
How do you access Claude Sonnet 5?
Use the model ID claude-sonnet-5 on the Claude API (or anthropic.claude-sonnet-5 on Amazon Bedrock, claude-sonnet-5 on Google Cloud). It is already the recommended Sonnet, and from 1 July it is the default for Free and Pro chat users. There is no waitlist.
Summary
Claude Sonnet 5 is Anthropic's new agent-focused mid-tier model: near-Opus 4.8 performance, a 1M-token context window, and introductory pricing of $2/$10 per million tokens through 31 August 2026. It out-performs Sonnet 4.6 across reasoning, tool use, and coding, and is safer than its predecessor, though Opus 4.8 still leads on the hardest tasks. For teams running agents at scale, it is the new default value pick.
