Anthropic · Claude Sonnet · Frontier
Claude Sonnet 4.5 pricing and limits
Anthropic's default model for agents and coding, and the model behind most of the loop incidents we analyze. The output rate is five times input, and explicit cache control decides whether long agent prompts are affordable.
| Line | Tokens | Rate / 1M | Cost |
|---|---|---|---|
| Input, uncached | 15,000 | $3.00 | $0.0450 |
| Input, cache read | 135,000 | $0.30 | $0.0405 |
| Output | 2,000 | $15.00 | $0.0300 |
- Input / 1M
- $3.00
- Output / 1M
- $15.00
- Cached / 1M
- $0.30
- Cache write / 1M
- $3.75
- Blended 3:1
- $6.00
- Context
- 200K
What drives Claude Sonnet 4.5 cost
Coding and browsing agents resend large contexts on every step: tool definitions, file contents, previous tool results. Without caching, that input dominates the bill. With cache breakpoints placed after the tools and system prompt, reads cost a tenth of input while the first write costs 1.25x for the five-minute TTL.
Sonnet 4.5 supports a 1M token context window in beta, and requests above 200K input tokens are billed at a higher long-context rate. Agents that never prune history drift into that band without anyone deciding to.
Cost of a typical call
Take a coding agent step with 150,000 input tokens, 135,000 of them served from cache, and 2,000 output tokens. At Claude Sonnet 4.5 list prices ($3.00 input, $0.30 cached input, $15.00 output per 1M) one call costs $0.1155, against $0.4800 with no cache hits. At 40,000 calls a month that is $4,620, or $19,200 uncached.
Blended at a 3:1 input to output ratio, Claude Sonnet 4.5 costs $6.00 per 1M tokens. Routing the same call to the fallback in the policy below, Claude Haiku 4.5, costs $0.0385, which is what a tenant pays you in COGS once they cross the hard cap and PALM SOUQ downgrades instead of blocking.
Cost comparison
Claude Sonnet 4.5 against three alternatives
The same coding agent step priced on Claude Haiku 4.5, GPT-5, Gemini 2.5 Pro. Move the sliders to match your traffic; models without a cache-read rate bill every input token in full.
Benchmark quality per task before switching. A cheaper model that needs more steps or longer outputs can cost more per finished task than its rate suggests, which is why PALM SOUQ reports cost per feature, not only per token.
Claude Sonnet 4.5 this guide
$0.1155 per call
$5,775 / mo
baseline
Claude Haiku 4.5
$0.0385 per call
$1,925 / mo
-67% vs Claude Sonnet 4.5
GPT-5
$0.0556 per call
$2,781 / mo
-52% vs Claude Sonnet 4.5
Gemini 2.5 Pro
$0.0806 per call
$4,030 / mo
-30% vs Claude Sonnet 4.5
List prices from the PALM SOUQ LLM Price Index, computed per call as uncached input + cache reads + output. Reasoning tokens count as output on reasoning models.
Rate limits by usage tier
RPM, input TPM and output TPM per model, raised with usage tier. Cached reads generally do not count toward input token limits, so improving cache hit rate is also a throughput fix.
Exact limits vary by account tier and change over time, so we do not reprint them here. Check the limits page for your own account in the Anthropic documentation at docs.anthropic.com. PALM SOUQ enforces your own per-tenant limits in front of the provider's, so a single customer cannot consume the shared allowance that every other tenant depends on.
Circuit-breaker best practices
Our worked coding-loop example: 2,500 calls of 150k input and 2k output at $3 and $15 per 1M cost $0.48 each, $1,200 in total over 45 minutes. A breaker that trips after 15 identical tool calls stops the same loop at $7.20.
Enable all four heuristics for Sonnet agents: tool thrashing for repeated calls, token velocity for parallel bursts, semantic echo for repeated plans, and a max step cap per session.
When a breaker trips, the gateway answers in under 50 ms with HTTP 429 and X-PalmSouq-Reason: circuit_breaker_triggered, before the request reaches Anthropic. Your agent framework sees a normal rate-limit response and stops, and the audit log records which heuristic fired and for which tenant.
# palmsouq.policy.yaml
policy: claude-sonnet-4-5-pro
match:
model: claude-sonnet-4.5
tier: pro # from X-PalmSouq-Tier
budgets:
monthly_usd: 500 # per tenant
soft_cap_pct: 80 # webhook to your app
on_hard_cap: fallback # or: block (HTTP 429)
fallback:
- claude-haiku-4.5
breakers:
tool_thrashing: { identical_calls: 3, window_steps: 8 }
token_velocity: { max_tokens_per_s: 50000, window_s: 10 }
semantic_echo: { repeats: 3 }
max_steps: 25
cache:
meter_reads: true
meter_writes: true
alert_read_share_below: 0.6- budgetsPer-tenant monthly dollar cap with an 80% soft-cap webhook, keyed by X-PalmSouq-Tenant.
- fallbackAt the hard cap, requests reroute to Claude Haiku 4.5 ($2.00 blended per 1M) instead of failing.
- breakersAll four loop heuristics on, evaluated in the request path with under 5 ms added latency.
Where it fits
Default for production coding agents, browser agents and multi-step workflows. Fall back to Haiku 4.5 for tenants over budget instead of blocking them.
Provider documentation: docs.anthropic.com
Claude Sonnet 4.5 pricing questions
How much does Claude Sonnet 4.5 cost per million tokens?
At list price Claude Sonnet 4.5 costs $3.00 per 1M input tokens and $15.00 per 1M output tokens, with cached input at $0.30 per 1M and five-minute cache writes at $3.75 per 1M. Prices were last reviewed in the PALM SOUQ LLM Price Index and are for the standard tier and shortest context band.
Does Claude Sonnet 4.5 support prompt caching?
Yes. Cache reads cost $0.30 per 1M tokens and cache writes cost $3.75 per 1M for the five-minute TTL. Caching saves money only when a prefix is read more often than it is rewritten, which is why PALM SOUQ meters reads and writes separately per tenant.
What are the Claude Sonnet 4.5 rate limits?
RPM, input TPM and output TPM per model, raised with usage tier. Cached reads generally do not count toward input token limits, so improving cache hit rate is also a throughput fix. For your account's exact numbers, see the Anthropic documentation at docs.anthropic.com.
How do I stop runaway Claude Sonnet 4.5 agent loops?
Our worked coding-loop example: 2,500 calls of 150k input and 2k output at $3 and $15 per 1M cost $0.48 each, $1,200 in total over 45 minutes. A breaker that trips after 15 identical tool calls stops the same loop at $7.20. The sample policy on this page enables all four PALM SOUQ heuristics with a max step cap of 25 and falls back to Claude Haiku 4.5 at the hard cap.
Is there a batch discount for Claude Sonnet 4.5?
50% off input and output through the Message Batches API. Batch traffic suits evaluations, backfills and nightly enrichment jobs that do not need a real-time answer.
Govern Claude Sonnet 4.5 spend per customer.
Point your Anthropic client at gateway.palmsouqltd.com, pass X-PalmSouq-Tenant, and every Claude Sonnet 4.5 token is attributed, capped and protected by circuit breakers.
14-day free production trial · 100k requests included · no card required