The Complete Guide to Claude Haiku 4.5
$1/$5 per MTok, 139,341 measured calls in 30 days (as of 2026-09-29) — third platform-wide
The cheapest, fastest tier in the Claude lineup at $1/$5 per million tokens; QCode measured 139,341 calls in 30 days (checked 2026-09-29) — third platform-wide, behind claude-sonnet-5 and deepseek-v4.1-flash.
Updated 2026-09-19
Claude, GPT and Chinese models on one key, billed per token; check /models for callable models.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard / AMEX), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Four key numbers
Per million tokens (input/output)
The cheapest Claude tier — a quarter of Opus 5.5 ($4/$20, 2026-09-22).
Real calls in the last 30 days
Measured 2026-09-29: Claude Haiku 4.5 logged 139,341 calls over 30 days on QCode, the third-highest volume, behind claude-sonnet-5 (202,293) and deepseek-v4.1-flash (179,461).
Platform-wide 30-day call volume rank (2026-09-29)
Measured 2026-09-29: Haiku 4.5's 30-day call volume ranks #3 among all models sold on QCode — one of the platform's workhorse tiers.
Positioning
Built for high-throughput, latency-sensitive, low-complexity tasks: classification, extraction, routing, simple edits.
What Haiku 4.5 is
Claude Haiku 4.5 is the cheapest, fastest tier in the Claude lineup, priced at $1/$5 per million tokens (input/output) — one-quarter the price of the current Opus 5.5 ($4/$20, released 2026-09-22). Its positioning isn't about reasoning depth, it's about throughput and response speed: classification, information extraction, request routing, formatting, and simple edits are its home turf — high-frequency, low-complexity work. Measured 30-day usage on QCode as of 2026-09-29 puts Haiku 4.5 at 139,341 calls, third platform-wide — a huge number of users have already made it a default choice for everyday high-frequency tasks.
Why its call volume is so high
Our own existing cost-optimization guidance already notes: use Sonnet for complex tasks and switch to Haiku for formatting and other simple work, avoiding Opus for anything Sonnet or Haiku can already handle. Haiku 4.5's low price and low latency make it a natural fit for being called heavily and repeatedly — routing decisions inside agentic workflows, bulk formatting, and similar tasks — which is exactly why its call volume ranks third platform-wide (measured 2026-09-29). It isn't used heavily on its own; it's continuously called by a huge number of workflows as the 'default lightweight tier.'
Timeline
Claude Haiku 4.5 remains on sale as the cheapest tier in the Claude lineup, priced at $1/$5 per MTok.
As agentic workflows spread, more and more frameworks set Haiku as the default lightweight routing tier.
Anthropic's Sonnet 5.5 launch post listed Claude Haiku 5.5 as part of the same family and gave its role line (high-volume and cost-sensitive applications); price, context window and model ID were still unpublished at the time. Released 2026-10-07, model ID claude-haiku-5-5.
Confirmed vs. needs your own verification
Confirmed
Haiku 4.5 is priced at $1/$5 per MTok; QCode shows 139,341 measured calls in 30 days (as of 2026-09-29, platform usage check); the specific routed snapshot ID is claude-haiku-4-5-20251001; the official deprecation list puts Haiku 4.5's retirement at "Not sooner than October 15, 2026" — the earliest possible day, not a shutdown date.
Needs your own verification
"Cheap models must be weak" is a common stereotype, but Haiku 4.5 performs reliably enough at what it's built for (classification, extraction, routing, simple edits). Whether it can handle your specific, more complex needs is best judged with a small-scale test, not by price alone.
Good fit for Haiku vs. needs a stronger tier
A good fit for Haiku 4.5
High throughput, latency-sensitive, and the task itself isn't complex: classification, extraction, routing, formatting, simple edits — Haiku offers the best value here.
Needs Sonnet or Opus instead
Complex reasoning, long agentic chains, deep code comprehension — Haiku's capability ceiling isn't built for this; step up to a higher tier.
How to use it
Call claude-haiku-4-5 directly on QCode, with the same interface as other Claude models. The recommended pattern is tiered routing: classify tasks by complexity first, route the simple, high-frequency work to Haiku, and route the complex, low-frequency work to Sonnet or Opus — controlling cost without sacrificing quality on harder tasks.
What to do on QCode
Haiku 4.5 shares one key with the whole Claude lineup on QCode, billed at official price × service rate; offloading high-frequency simple tasks to Haiku is one of the most direct, effective ways to control your overall token budget.
FAQ
Where does Haiku 4.5 rank by platform-wide call volume?
On measured 30-day usage as of 2026-09-29, Haiku 4.5 ranks third with 139,341 calls, behind claude-sonnet-5 (202,293) and deepseek-v4.1-flash (179,461). The ranking moves with usage.
How much cheaper is Haiku 4.5 than Sonnet 5 or the Opus series?
Haiku 4.5 is $1/$5 per MTok, Sonnet 5 is $2/$10, the current Claude Opus 5.5 (released 2026-09-22) is $4/$20, the previous Opus 5 stays at $5/$25, and Fable 5.1 is $10/$50 (Anthropic's official price list, checked 2026-09-29) — so Haiku is half of Sonnet 5 and a quarter of Opus 5.5. Anthropic made Sonnet 5's $2/$10 permanent on 2026-08-10, so the planned 2026-09-01 move to $3/$15 no longer applies.
What role does Haiku 4.5 typically play in agentic setups?
It's well-suited for routing decisions, extracting tool-call parameters, and formatting results — high-frequency subtasks that don't need deep reasoning, leaving complex decisions to Sonnet or Opus.
Why does it have such high call volume when many people don't realize they're using it?
Many frameworks/workflows set Haiku as the default lightweight tier, automatically handling classification, routing, and other background tasks. Users often only notice the main model (like Sonnet), while Haiku gets called heavily behind the scenes.
How does Haiku 4.5 relate to Claude Code's default model?
Claude Code defaults to Sonnet 5; Haiku 4.5 is more of an optional lightweight tier that users or frameworks explicitly route simple tasks to, rather than the default main interaction model.
How is Haiku 4.5 billed on QCode?
At the official rate of $1/$5 per million tokens (input/output) × service rate, matching Anthropic's own API pricing baseline, with no additional markup.
Sources
Haiku 4.5's pricing and positioning come from an already-verified Claude cost-optimization page on this site; the 30-day real usage figure comes from QCode platform usage verification, pulled 2026-09-29; the cited Anthropic price list, deprecation list and 2026-09-28 post come from archives checked the same day (routed snapshot ID: claude-haiku-4-5-20251001).
Hand off high-frequency tasks to Haiku 4.5
One QCode key calls Haiku 4.5 and the whole Claude lineup, billed at official price × service rate.
Related reading
Claude's 5-Hour Rolling Window Explained
The quota mechanism that also matters when calling Haiku at high frequency.
The complete QCode.cc pricing guide
The 7 plan tiers and how billing works.
Why sub-agents can burn through your quota in 30 minutes
How to sensibly split usage between Haiku and higher-tier models in multi-agent workflows.
This page's model pricing and usage data are based on already-verified information on this site and QCode's platform usage verification, and are not an official Anthropic statement; actual call volume and sale status are governed by live QCode platform data.