llmrelay
// claude

Claude Haiku 4.5

The fastest and cheapest Claude. Perfect for classification and routing.

Claude Haiku 4.5 is the small, fast model in the Claude line - built for the kind of work you run millions of times rather than once. Classification, routing, guardrails, batch tagging, input validation: tasks where the answer is short and speed matters more than depth.

Anthropic rates its comparative latency Very Fast, and that is the point. A model that returns in under 500ms can sit in a request path without the user feeling it, which is what lets you put an LLM in front of every API call or every message in a moderation queue.

We sell it at half Anthropic’s list price - $0.50 per million input tokens against $1.00. At that rate the model cost stops being the budget constraint for high-volume work.

Official list
llmrelay
Input / M tokens
$1.00
$0.50
Output / M tokens
$5.00
$2.50
Context window
—
200,000 tokens
Max output
—
64,000 tokens

What it costs you per month

Real-world budget scenarios. Numbers are simple sums — official list price vs llmrelay's 50% off tier.

Usage scenario
Official
llmrelay
Light coding (1M in / 200K out per month)
$2.00
$1.00save $1.00
Heavy Cursor / Cline user (50M in / 5M out per month)
$75.00
$37.50save $37.50
Production RAG (500M in / 20M out per month)
$600.00
$300.00save $300.00

What it's good at

Best for

Pick Claude Haiku 4.5 when

  • +Classification or routing at scale. Intent detection, sentiment tagging, spam filtering, category assignment - Haiku is fast enough and cheap enough to run on every request.
  • +Guardrails and input validation. Check user input for prompt injection, PII leakage, policy violations before handing it to a more expensive model downstream.
  • +Batch tagging jobs where volume matters more than nuance. Labelling a million documents is a Haiku job, not an Opus job.
  • +You need a Claude model but the work is simple. Haiku shares the same API shape and safety tuning as Opus and Sonnet, so swapping it in is one model-id change.

Choose something else when

  • !Long-context work. The 200k context window is small for the Claude family - Sonnet 5 and Opus 5 carry 1M.
  • !Complex reasoning or multi-step agent work. Haiku is optimised for speed, not depth. If you need to think through a problem rather than label it, reach for Sonnet or Opus.
  • !Creative or narrative writing. That is what Fable 5 is for.
  • !Work where the model needs to synthesise information across many sources. Haiku’s shallow context and tuning make it the wrong tool for research or analysis.

Questions people ask about Claude Haiku 4.5

How much does Claude Haiku 4.5 cost?

Anthropic’s list price is $1.00 per million input tokens and $5.00 per million output tokens. On llmrelay it is $0.50 and $2.50 - exactly half list. Billing is prepaid per token from a single credit pool, with no subscription or volume commit.

Is Haiku 4.5 good enough for real work?

For the right work, yes. It is demonstrably good at classification, sentiment analysis, short-form Q&A, and routing decisions. It is not good at reasoning through ambiguity or synthesising across long context. The question is not whether Haiku is capable in some absolute sense - it is whether your specific task needs depth or speed.

What is the context window on Claude Haiku 4.5?

200,000 tokens input, with up to 64,000 output tokens. That is smaller than Sonnet 5 and Opus 5 (both 1M), so if you are putting entire codebases or long documents in front of the model, Haiku is the wrong tier.

Can I use Haiku 4.5 with Claude Code, Cursor or Cline?

Technically yes - it speaks the same API - but it is the wrong model for agentic coding. Cursor and Cline lean on long context and multi-step reasoning, which is where Haiku falls short. Use Sonnet 5 or Opus 5 for that workload.

Try Claude Haiku 4.5 at half the price

Free to create an account, no subscription, $10 minimum top-up — enough to run this model against a real task and judge quality yourself.

Get API key →

Compare Claude Haiku 4.5 against the alternatives

The comparisons this model appears in, the models nearest it on price, and the full rate card.