Anthropic · Claude Haiku · Small
Claude 3.5 Haiku pricing and limits
The previous small Claude model, slightly cheaper than Haiku 4.5. Still used for high-volume classification and as a final fallback tier.
| Line | Tokens | Rate / 1M | Cost |
|---|---|---|---|
| Input, uncached | 1,000 | $0.80 | $0.00080 |
| Input, cache read | 4,000 | $0.080 | $0.00032 |
| Output | 200 | $4.00 | $0.00080 |
- Input / 1M
- $0.80
- Output / 1M
- $4.00
- Cached / 1M
- $0.080
- Cache write / 1M
- $1.00
- Blended 3:1
- $1.60
- Context
- 200K
What drives Claude 3.5 Haiku cost
Cost is driven by volume and prompt size. Cache writes at 1.25x input and reads at a tenth of input apply as on other Claude models, which matters when thousands of classification calls share a long label taxonomy in the prompt.
It has no extended thinking, so output cost tracks visible output. That makes it predictable, and a good choice when you need a hard per-call cost bound.
Cost of a typical call
Take a classification call with 5,000 input tokens, 4,000 of them served from cache, and 200 output tokens. At Claude 3.5 Haiku list prices ($0.80 input, $0.080 cached input, $4.00 output per 1M) one call costs $0.00192, against $0.00480 with no cache hits. At 1,500,000 calls a month that is $2,880, or $7,200 uncached.
Blended at a 3:1 input to output ratio, Claude 3.5 Haiku costs $1.60 per 1M tokens. Routing the same call to the fallback in the policy below, Claude 3.5 Haiku, costs $0.00192, 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 3.5 Haiku against three alternatives
The same classification call priced on Claude Haiku 4.5, GPT-4o mini, Gemini 2.5 Flash-Lite. 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 3.5 Haiku this guide
$0.00192 per call
$1,920 / mo
baseline
Claude Haiku 4.5
$0.00240 per call
$2,400 / mo
+25% vs Claude 3.5 Haiku
GPT-4o mini
$0.00057 per call
$570 / mo
-70% vs Claude 3.5 Haiku
Gemini 2.5 Flash-Lite
$0.00028 per call
$280 / mo
-85% vs Claude 3.5 Haiku
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 and token per minute limits per model by usage tier. Limits are separate from Haiku 4.5, so splitting traffic across both can add headroom during migrations.
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
Use it as the last step of a fallback chain with a hard monthly cap behind it. When a tenant exhausts even the fallback budget, return HTTP 429 with a clear reason header so your app can show an upgrade prompt.
Schema validation retries are the common loop here; cap retries per task at the gateway.
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-3-5-haiku-free
match:
model: claude-3.5-haiku
tier: free # from X-PalmSouq-Tier
budgets:
monthly_usd: 20 # 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: 6
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
Bulk classification, tagging, short summaries and last-resort fallback for budget-capped tenants.
Provider documentation: docs.anthropic.com
Claude 3.5 Haiku pricing questions
How much does Claude 3.5 Haiku cost per million tokens?
At list price Claude 3.5 Haiku costs $0.80 per 1M input tokens and $4.00 per 1M output tokens, with cached input at $0.080 per 1M and five-minute cache writes at $1.00 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 3.5 Haiku support prompt caching?
Yes. Cache reads cost $0.080 per 1M tokens and cache writes cost $1.00 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 3.5 Haiku rate limits?
RPM and token per minute limits per model by usage tier. Limits are separate from Haiku 4.5, so splitting traffic across both can add headroom during migrations. For your account's exact numbers, see the Anthropic documentation at docs.anthropic.com.
How do I stop runaway Claude 3.5 Haiku agent loops?
Use it as the last step of a fallback chain with a hard monthly cap behind it. When a tenant exhausts even the fallback budget, return HTTP 429 with a clear reason header so your app can show an upgrade prompt. The sample policy on this page enables all four PALM SOUQ heuristics with a max step cap of 6 and falls back to Claude 3.5 Haiku at the hard cap.
Is there a batch discount for Claude 3.5 Haiku?
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 3.5 Haiku spend per customer.
Point your Anthropic client at gateway.palmsouqltd.com, pass X-PalmSouq-Tenant, and every Claude 3.5 Haiku token is attributed, capped and protected by circuit breakers.
14-day free production trial · 100k requests included · no card required