How does Sonnet 5 differ from Sonnet 5.5?

Anthropic lists five breaking changes: turning off up-front thinking uses between_tools (valid only at high effort or below); forced tool use returns a 400; thinking blocks are bound to the model and conversation; computer use on the Claude API accepts only computer_toolset_20260801; and the advisor tool no longer accepts Opus 4.8, Opus 4.7, Sonnet 5 or Haiku 5.5 as advisors. 5.5 also recalibrates effort, moves longer text between tool calls into thinking blocks, and lowers the minimum cacheable prompt from 1,024 to 512 tokens; Anthropic says its output is more than 30% faster. The tokenizer is the same.

When should I keep Sonnet 5 pinned?

When an integration depends on disabling thinking (especially at xhigh or max), forced tool_choice, computer_20251124, or a UI that streams the text blocks between tool calls, and cannot change yet. Anthropic commits to retiring Sonnet 5 no sooner than June 30, 2027.

When should I move up from Sonnet 5 to Opus?

When a task fails for reasoning reasons rather than running out of room. Sonnet 5 and every Opus generation on sale carry a 1M-token context window, so context size is not what separates them.

Claude Sonnet 5

Save 72%Official

anthropic/claude-sonnet-5

The previous Sonnet — succeeded by Sonnet 5.5.

Get an API key View code sample

Live model playground

Try Claude Sonnet 5 here

Tune the request, see the current price ceiling, and inspect the real response without leaving this model page.

streamchat

1Configure input

Prompt

97 / 20,000

Add images

Maximum estimated cost—

2Live resultExample

Claude Sonnet 5

HTTP 429 Too Many Requests means the client has exceeded the rate limit set by the server. The server will typically include a Retry-After header indicating how many seconds to wait before trying again. A well-behaved client should pause for that duration (or use exponential backoff with jitter if no header is present), then retry the request. Persistent 429s despite backing off usually signal that the client needs to reduce its overall request rate or request a higher quota from the API provider.

Context window: 1M
Max output: 128K
Input modalities: TextImage
Output modalities: Text
Capabilities: Vision · Reasoning · Tool calling · Coding · Prompt cache · Paid cache writes
Upstream release: 2026-06-30
RouteMux name: anthropic/claude-sonnet-5

What is Claude Sonnet 5

Claude Sonnet 5 was released by Anthropic on June 30, 2026, succeeding Sonnet 4.6, and became legacy when Sonnet 5.5 arrived on September 28, 2026, with retirement not sooner than June 30, 2027. It has a 1M-token context window, up to 128K output tokens per request, image input and tool calling, and a reliable knowledge cutoff of January 2026. At launch Anthropic called it its most agentic Sonnet yet: it plans, uses tools such as browsers and terminals, runs autonomously, and performs close to Opus 4.8.

Against Sonnet 4.6 it was a generational change: adaptive thinking on by default, budget_tokens and non-default sampling parameters returning a 400, and the new tokenizer, which produces roughly 30% more tokens for the same text. Anthropic's reference point is that Sonnet 5 at medium effort is comparable to Sonnet 4.6 at high.

Against Sonnet 5.5 it keeps several request shapes that 5.5 no longer accepts: thinking "disabled" at any effort level; forced tool use (tool_choice "any" or a named tool); computer_20251124 on the Claude API; and text written between tool calls returned as text blocks. An integration that relies on these and cannot change yet can stay pinned to Sonnet 5; otherwise Anthropic recommends moving to Sonnet 5.5. The two share a tokenizer, so token counts do not change when you switch.

Through RouteMux, Sonnet 5 costs $0.56 per 1M input tokens against Anthropic's $2, and $2.8 per 1M output tokens against $10.

Before you choose

How does Sonnet 5 differ from Sonnet 5.5?: Anthropic lists five breaking changes: turning off up-front thinking uses between_tools (valid only at high effort or below); forced tool use returns a 400; thinking blocks are bound to the model and conversation; computer use on the Claude API accepts only computer_toolset_20260801; and the advisor tool no longer accepts Opus 4.8, Opus 4.7, Sonnet 5 or Haiku 5.5 as advisors. 5.5 also recalibrates effort, moves longer text between tool calls into thinking blocks, and lowers the minimum cacheable prompt from 1,024 to 512 tokens; Anthropic says its output is more than 30% faster. The tokenizer is the same.
When should I keep Sonnet 5 pinned?: When an integration depends on disabling thinking (especially at xhigh or max), forced tool_choice, computer_20251124, or a UI that streams the text blocks between tool calls, and cannot change yet. Anthropic commits to retiring Sonnet 5 no sooner than June 30, 2027.
When should I move up from Sonnet 5 to Opus?: When a task fails for reasoning reasons rather than running out of room. Sonnet 5 and every Opus generation on sale carry a 1M-token context window, so context size is not what separates them.

Claude Sonnet 5 API pricing

Prices verified 2026-10-06

InputOfficial $2 / M tokens

$0.56/ M tokensSave 72%

OutputOfficial $10 / M tokens

$2.8/ M tokensSave 72%

Cache read $0.056/M · Cache write $0.7/M

Estimate cost

Input tokensOutput tokens

Estimated cost$0.0084

Key capabilities

Thinking can be disabled at any effort: Thinking is on by default, but thinking "disabled" is accepted at every effort level. Sonnet 5.5 only offers between_tools, and only at high or below.
Forced tool use still works: tool_choice "any" or a named tool is honored. Sonnet 5.5 returns a 400.
Same tokenizer as Sonnet 5.5: Token counts stay the same when moving to 5.5. Coming from 4.6, the same text produces roughly 30% more tokens.
Sampling parameters dropped before forwarding: Sonnet 5 rejects non-default temperature, top_p and top_k. RouteMux strips them before forwarding, so SDKs that send them by default do not fail, though the values have no effect.

Limited time10% extra Credits on your first top-up — up to 200 bonus Credits

Applies to your first top-up within 72 hours of signing up. Minimum $20, once per account.

Interface and protocols

Anthropic MessagesStreaming

Use any of these IDs to call this model via the API.

RouteMux nameanthropic/claude-sonnet-5

Replace the ROUTEMUX_KEY placeholder with your API key. Create one →

from anthropic import Anthropic

client = Anthropic(
    base_url="https://api.routemux.com/anthropic",
    api_key="<ROUTEMUX_KEY>",
)

message = client.messages.create(
    model="anthropic/claude-sonnet-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "What is the meaning of life?"}],
)
print(message.content[0].text)

base_url: https://api.routemux.com/anthropic

Performance (last 30 days)

2,688 requests · last 30 days

Total usage28.93M

P50 first-token latency5.97s

Avg throughput27 tok/s

Success rate100.0%

Uptime · 24htested via anthropic_messages

100.00%

Success rate · 60d · from real requests

Claude Sonnet 5 FAQ

›What is Claude Sonnet 5?

The previous Sonnet — succeeded by Sonnet 5.5.

›How large is the context window of Claude Sonnet 5?

Claude Sonnet 5 supports a context window of up to 1M tokens, with up to 128K output tokens per request.

›How much does the Claude Sonnet 5 API cost?

Through RouteMux, Claude Sonnet 5 costs $0.56 for input and $2.8 for output (USD per 1M tokens), below the official list price.

›How do I call Claude Sonnet 5 via API?

Claude Sonnet 5 is OpenAI-compatible: point your base URL at RouteMux and set the model field to anthropic/claude-sonnet-5 — no code changes needed.

›Which input and output modalities does Claude Sonnet 5 support?

Claude Sonnet 5 accepts text, image as input and produces text as output.

›When was Claude Sonnet 5 released?

Claude Sonnet 5 was released on June 30, 2026.

Sources

Reviewed by the RouteMux team on 2026-10-24

View as Markdown