Anthropic · Claude Haiku · Small
Claude Haiku 4.5 pricing and limits
Anthropic's fast small model with near-frontier coding ability. A natural fallback target for Sonnet and a strong default for sub-agents.
| Line | Tokens | Rate / 1M | Cost |
|---|---|---|---|
| Input, uncached | 2,000 | $1.00 | $0.00200 |
| Input, cache read | 8,000 | $0.10 | $0.00080 |
| Output | 700 | $5.00 | $0.00350 |
- Input / 1M
- $1.00
- Output / 1M
- $5.00
- Cached / 1M
- $0.10
- Cache write / 1M
- $1.25
- Blended 3:1
- $2.00
- Context
- 200K
What drives Claude Haiku 4.5 cost
Haiku 4.5 is cheap enough to run as the worker in orchestrator and sub-agent designs, which is where fan-out gets expensive: one planner request can spawn dozens of Haiku calls, each resending tool schemas. Caching those schemas cuts input to a tenth of the list rate after the first write.
Extended thinking is available and billed as output. For sub-agent tasks that only need a lookup or short transformation, leaving thinking off keeps each call close to its visible output size.
Cost of a typical call
Take a sub-agent call with 10,000 input tokens, 8,000 of them served from cache, and 700 output tokens. At Claude Haiku 4.5 list prices ($1.00 input, $0.10 cached input, $5.00 output per 1M) one call costs $0.00630, against $0.0135 with no cache hits. At 500,000 calls a month that is $3,150, or $6,750 uncached.
Blended at a 3:1 input to output ratio, Claude Haiku 4.5 costs $2.00 per 1M tokens. Routing the same call to the fallback in the policy below, Claude 3.5 Haiku, costs $0.00504, 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 Haiku 4.5 against three alternatives
The same sub-agent call priced on Claude 3.5 Haiku, GPT-5 mini, Gemini 2.5 Flash. 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 Haiku 4.5 this guide
$0.00630 per call
$3,150 / mo
baseline
Claude 3.5 Haiku
$0.00504 per call
$2,520 / mo
-20% vs Claude Haiku 4.5
GPT-5 mini
$0.00210 per call
$1,050 / mo
-67% vs Claude Haiku 4.5
Gemini 2.5 Flash
$0.00295 per call
$1,475 / mo
-53% vs Claude Haiku 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
Per-model RPM and input and output TPM by usage tier. Small models usually get higher ceilings, but parallel sub-agents from one tenant can still saturate them.
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
Cap concurrent sub-agents per tenant and per session. PALM SOUQ counts in-flight requests by tenant and feature header, which catches runaway fan-out before the provider does.
When used as a fallback, apply the same breakers as the primary model so a tenant downgraded at 90% of budget cannot loop on the cheaper model instead.
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-haiku-4-5-pro
match:
model: claude-haiku-4.5
tier: pro # from X-PalmSouq-Tier
budgets:
monthly_usd: 150 # per tenant
soft_cap_pct: 80 # webhook to your app
on_hard_cap: fallback # or: block (HTTP 429)
fallback:
- claude-3.5-haiku
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: 15
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 3.5 Haiku ($1.60 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
Sub-agents, classification, routing, support replies and as the downgrade target when a Sonnet tenant nears its cap.
Provider documentation: docs.anthropic.com
Claude Haiku 4.5 pricing questions
How much does Claude Haiku 4.5 cost per million tokens?
At list price Claude Haiku 4.5 costs $1.00 per 1M input tokens and $5.00 per 1M output tokens, with cached input at $0.10 per 1M and five-minute cache writes at $1.25 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 Haiku 4.5 support prompt caching?
Yes. Cache reads cost $0.10 per 1M tokens and cache writes cost $1.25 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 Haiku 4.5 rate limits?
Per-model RPM and input and output TPM by usage tier. Small models usually get higher ceilings, but parallel sub-agents from one tenant can still saturate them. For your account's exact numbers, see the Anthropic documentation at docs.anthropic.com.
How do I stop runaway Claude Haiku 4.5 agent loops?
Cap concurrent sub-agents per tenant and per session. PALM SOUQ counts in-flight requests by tenant and feature header, which catches runaway fan-out before the provider does. The sample policy on this page enables all four PALM SOUQ heuristics with a max step cap of 15 and falls back to Claude 3.5 Haiku at the hard cap.
Is there a batch discount for Claude Haiku 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 Haiku 4.5 spend per customer.
Point your Anthropic client at gateway.palmsouqltd.com, pass X-PalmSouq-Tenant, and every Claude Haiku 4.5 token is attributed, capped and protected by circuit breakers.
14-day free production trial · 100k requests included · no card required