What is Claude Sonnet 5?
Sonnet 5 is Anthropic's current Sonnet model and a drop-in upgrade for Sonnet 4.6. It reads text and images, writes text, and works across a 1M-token context with up to 128K tokens of output; its knowledge stops at January 2026. Thinking is adaptive and on by default, and it runs on the Claude API, Amazon Bedrock, Google Cloud and Microsoft Foundry.
It sits below Opus 5.5 in Anthropic's line-up. Anthropic now points most API work to Opus 5.5 first, while Sonnet 5 answers faster at half the token price. A new tokenizer means the same text counts as more tokens than it did on Sonnet 4.6.
What changed from Sonnet 4.6?
It is cheaper per token and scores higher on Anthropic's tests, but it thinks by default, counts more tokens and rejects some old API settings.
- Price
- $2 / $10 per million tokens, down from $3 / $15.
- Tokenizer
- The same text becomes about 30% more tokens, so real savings are smaller than the price cut.
- Thinking
- Adaptive thinking is on by default, and manual thinking budgets now return an error.
- Sampling
- Custom temperature, top_p and top_k values are rejected.
- Tools
- Adds the browser use tool and the newer computer use toolset on the Claude API and Google Cloud.
- Safeguards
- It is the first Sonnet with real-time cyber safeguards, which can refuse high-risk security requests.
Who is Claude Sonnet 5 best for?
Claude Sonnet 5 is best for developers running agents and coding tools on a budget, work that needs a very long context, and anyone who wants a strong Claude model for free.
Agents on a budget
It plans, drives browsers and terminals and works through long tasks close to Opus 4.8 level on Anthropic's tests, at $2 / $10 per million tokens against Opus 4.8's $5 / $25.
Coding and debugging
Anthropic built it for multi-step software work such as tracing a bug, writing a test and shipping the fix, and it scores 63.2% on SWE-bench Pro against 58.1% for Sonnet 4.6.
Large codebases and documents
The full 1M-token context costs the same per token as a short prompt, so long inputs carry no surcharge.
Trying Claude for free
Claude's Free plan includes Sonnet on web, desktop and mobile, so you can test it before paying for Pro or the API.
What's included with Claude Sonnet 5?
Five effort levels, adaptive thinking, browser and computer use tools, half-price batch jobs, cheap cached reads, and every major cloud.
- Five effort levels
- Low, medium, high, xhigh and max trade answer depth against speed and cost; high is the default.
- Adaptive thinking
- On by default, so the model decides how much to think per request, and you can switch it off for plain replies.
- Browser and computer use
- On the Claude API and Google Cloud it supports the browser use tool and the newer computer use toolset, which Sonnet 4.6 lacks.
- Batch API
- Half-price jobs at $1 / $5 per million tokens, with up to 300K output tokens under a beta header.
- Prompt caching
- Cache reads cost $0.20 per million tokens, a tenth of the normal input price.
- Major clouds
- Runs on the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS, with zero data retention for eligible organizations.
How much does Claude Sonnet 5 cost?
Claude Sonnet 5 costs $2 per million input tokens and $10 per million output tokens on the API, and it is included in Claude's Free plan, with Pro at $20 a month and Max from $100 a month for more usage.
- Sonnet in chat on web, desktop and mobile
- Web search, file creation and code runs
- Usage limits reset every five hours
- Cache reads $0.20 and 5-minute cache writes $2.50 per million tokens
- Batch processing halves it to $1 / $5
- Full 1M-token context at the same rate
- At least 5x the Free plan's usage per session
- Claude Code included
- Opus models as well as Sonnet
- 5x or 20x the usage of Pro
- Higher output limits
- Priority access at busy times
Pro is $17 a month when paid yearly ($200 up front); Max is billed monthly only. Team seats cost $25 a month (Standard) or $125 (Premium), or $20 and $100 billed yearly, and Enterprise is $20 per seat a month billed yearly, plus usage at API rates. Every plan has usage limits that reset on a rolling five-hour window. On the API, 1-hour cache writes cost $4 per million tokens and US-only inference costs 1.1 times the standard rate. Prices are in US dollars. Full pricing on claude.com. We checked these prices on .
How do readers rate Claude Sonnet 5?
No one has rated Claude Sonnet 5 yet. Be the first.
Sign in to rate. One rating per account.
Claude Sonnet 5 pros and cons
It comes close to Opus 4.8 for 40% of the price and is free in the apps, but it counts more tokens per text and trails Opus on the hardest coding work.
Pros
- At $2 / $10 per million tokens it costs half as much as Opus 5.5 and less than Sonnet 4.6, and Hacker News developers called it a good low-cost workhorse at that price, especially at low effort.
- Beats Sonnet 4.6 on every benchmark in Anthropic's launch table, including 80.4% against 67.0% on Terminal-Bench 2.1, and edges Opus 4.8 on GDPval-AA v2 knowledge work, 1618 to 1615.
- Included on Claude's Free plan, where Opus models are not, as well as on every paid plan.
- Unlike Opus 5.5, it is in Anthropic's Cyber Verification Program, so verified security teams can apply to have dual-use blocks relaxed.
Cons
- It uses more tokens than the price suggests: the new tokenizer turns the same text into about 30% more tokens than Sonnet 4.6, and several Hacker News users say it spends more than they expect, one burning a whole five-hour quota on a single research run.
- Several Hacker News developers found it clearly weaker than Opus for hard software work, and some said it skipped steps or ignored explicit instructions.
- Since Opus 5.5 arrived at $4 / $20 with newer knowledge, several Reddit users say they have stopped using Sonnet 5 because Opus 5.5 gets more done for the money.
- Code written for Sonnet 4.6 can break: custom temperature, top_p or top_k values and manual thinking budgets return errors, and Priority Tier is not available.
What are the best Claude Sonnet 5 alternatives?
Six tools on Recaply do Claude Sonnet 5's job. Command A+, GLM-4.7 and LongCat-2.0 are open source, so you can run them on your own server.
Claude Opus 5.5PaidChoose it when quality matters more than cost; Anthropic now suggests it as the starting point for most API work, it costs $4 / $20 per million tokens, and its knowledge runs to June 2026.
Gemini 3.8 FlashFreeChoose it when you need video, audio or PDF input or a free API tier; its standard rate is $1.50 / $7.50 per million tokens, though output stops at about 65K tokens.
Kimi K2.7 CodeFreeChoose it when you want downloadable weights under a modified MIT licence and a lower bill of $0.95 / $4 per million tokens, and a 256K context is enough.
Command A+Open SourceChoose it when agent workloads must run inside your own environment, with free downloadable weights instead of a hosted API.
GLM-4.7Open SourceChoose it when you want an MIT-licensed agentic coding model with a 200K context that you can self-host.
LongCat-2.0Open SourceChoose it when you want an MIT-licensed 1.6T MoE model with a 1M-token context for coding that you can self-host.
Questions about Claude Sonnet 5
What is Claude Sonnet 5?
Claude Sonnet 5 is Anthropic's fast mid-tier model for coding, agents and everyday chat, with a 1M-token context, API pricing at half of Opus 5.5, and a place on Claude's Free plan.
Who is Claude Sonnet 5 for?
Claude Sonnet 5 is best for developers running agents and coding tools on a budget, work that needs a very long context, and anyone who wants a strong Claude model for free.
Is Claude Sonnet 5 free?
Yes, in the Claude apps. Claude's Free plan includes Sonnet on web, desktop and mobile, with usage limits that reset every five hours. Developers who call Claude Sonnet 5 through the API pay $2 per million input tokens and $10 per million output tokens.
How much does Claude Sonnet 5 cost?
Claude Sonnet 5's paid plans are API at $2 / $10 per million input / output tokens, Pro at $20 per month and Max at From $100 per month.
Is there an open source alternative to Claude Sonnet 5?
Yes. Command A+, GLM-4.7 and LongCat-2.0 are open source tools that do the same job, so you can run your own.
What are the best Claude Sonnet 5 alternatives?
The closest alternatives on Recaply are Claude Opus 5.5, Gemini 3.8 Flash, Kimi K2.7 Code, Command A+, GLM-4.7 and LongCat-2.0.
Where can I try Claude Sonnet 5?
On its official website, anthropic.com.
Does Claude Sonnet 5 use more tokens than Sonnet 4.6?
Yes. Claude Sonnet 5 uses a new tokenizer, and Anthropic says the same text comes out as about 30% more tokens, depending on the content. Its rate of $2 / $10 per million tokens still undercuts Sonnet 4.6's $3 / $15, but an equivalent request does not get cheaper in proportion.
Should I use Claude Sonnet 5 or Opus 5.5?
Anthropic now suggests starting with Opus 5.5 for most API work. Pick Sonnet 5 when speed and token cost matter most, or when you want it free in the Claude apps; Opus 5.5 costs twice as much per token, is not on the Free plan, and its knowledge runs to June 2026 rather than January 2026.
What changes when I move from Sonnet 4.6 to Sonnet 5 on the API?
Change the model ID to claude-sonnet-5, remove non-default temperature, top_p and top_k values, and replace thinking budgets with adaptive thinking and the effort setting. Thinking is on by default, so read replies by block type, and recount tokens and max_tokens under the new tokenizer.
Can I use Claude Sonnet 5 for security work?
For routine defensive work, yes, but real-time safeguards block high-risk cyber requests by default. Verified security teams can apply to Anthropic's Cyber Verification Program, which covers Sonnet 5 but not yet Opus 5.5; Anthropic still recommends Opus 4.8 for work that needs looser guardrails.
Our verdict
Sonnet 5 is the Claude model to pick when speed and token cost matter more than peak quality, and the one you get free in the Claude apps.
It beats Sonnet 4.6 on every test Anthropic published and comes close to Opus 4.8, for $2 / $10 per million tokens with the full 1M context.
- Skip it for
- Long, hard coding jobs, where many developers now find Opus 5.5 better value at twice the token price.
- Check first
- Recount your prompts under the new tokenizer before you budget, since the same text uses about 30% more tokens.
Claude Sonnet 4.6
Claude Fable 5
Claude Opus 4.7
SubQ