Claude Haiku 5.5 API: Pricing, the 100K Tier and Moving from Haiku 4.5

Claude Haiku 5.5 is the newest Haiku model, released by Anthropic on October 7, 2026. On Onysoft AI Gateway it is available as anthropic/claude-haiku-5.5. As of October 17, 2026, for prompts up to 100,000 tokens the list price is $0.15 per 1M input tokens (≈7.41 TL) and $0.75 per 1M output tokens (≈37.07 TL); for prompts above 100,000 tokens unit prices rise fivefold ($0.75 / $3.75). The context window is 1,000,000 tokens.

This guide covers how Anthropic positions Haiku 5.5, the USD and TL price table on Onysoft, how the 100K tier affects your bill, what to watch when moving from Haiku 4.5, a sub-agent cost scenario with Opus 5.5 and Sonnet 5.5, and working Python/curl examples.

What Is Claude Haiku 5.5? What Anthropic Announced

Anthropic presents Haiku 5.5 as the cheapest, fastest and most capable small model it has released so far. It is positioned for high-volume, cost-sensitive work such as summaries, context compaction, database queries and classification; for speed-sensitive tasks such as live customer support and browser use; and as a sub-agent alongside Opus 5.5 or Sonnet 5.5 on coding work. According to Anthropic, Haiku 5.5 does best on narrowly scoped tasks, while Sonnet 5.5 and Opus 5.5 remain the better choices for complex agentic coding.

Item	Detail
Release date	October 7, 2026 (Anthropic announcement)
Anthropic API ID	claude-haiku-5-5
Onysoft IDs	anthropic/claude-haiku-5.5, ~anthropic/claude-haiku-latest
Context	1,000,000 tokens (200,000 on Haiku 4.5)
Max output	128,000 tokens (Anthropic model documentation)
Price tier	All unit prices 5x when the prompt exceeds 100,000 tokens

Benchmark results published by Anthropic (October 7, 2026 announcement; Onysoft has not reproduced these measurements independently):

Benchmark	Claude Haiku 5.5	Claude Haiku 4.5
GDPval-AA v2.1 (Elo)	1620	735
OSWorld 2.1 (offline subset)	72.4%	15.7%
Humanity's Last Exam (no tools)	45.9%	10.2%
Terminal-Bench 4.0	39.2%	0.0%

Anthropic also says Haiku 5.5 costs around 75% less to run than Haiku 4.5 on average; that is a workload-dependent estimate, separate from the list-price difference. Use the Playground to measure it on your own prompts.

Claude Haiku 5.5 Pricing: USD and TL Table

The table shows Haiku 5.5 and nearby models at Onysoft list prices as of October 17, 2026, per 1 million tokens:

Model	API ID	Context (tokens)	1M input	1M output	1M input (TL)	1M output (TL)
Claude Haiku 5.5 (prompt ≤100K)	anthropic/claude-haiku-5.5	1,000,000	$0.15	$0.75	7.41 TL	37.07 TL
Claude Haiku 5.5 (prompt >100K)	anthropic/claude-haiku-5.5	1,000,000	$0.75	$3.75	37.07 TL	185.33 TL
Claude Haiku 4.5	anthropic/claude-haiku-4.5	200,000	$1.50	$7.50	74.13 TL	370.65 TL
Claude Sonnet 5.5	anthropic/claude-sonnet-5.5	1,000,000	$3.00	$15.00	148.26 TL	741.31 TL
Claude Opus 5.5	anthropic/claude-opus-5.5	1,000,000	$6.00	$30.00	296.52 TL	1,482.61 TL
GPT-6 Luna	openai/gpt-6-luna	1,050,000	$0.15	$0.75	7.41 TL	37.07 TL

Prices are Onysoft list prices as of October 17, 2026; see /models for current pricing. Lira figures use 1 USD = 49.4204 TL (TCMB rate of October 16, 2026); billing uses the TCMB rate at request time. On GPT-6 Luna, prompts reaching 272,000 tokens are billed at 2x input and 1.5x output.

What the table shows:

• 90% lower list price than Haiku 4.5: for prompts up to 100,000 tokens, $0.15 instead of $1.50 per 1M input and $0.75 instead of $7.50 per 1M output. For longer prompts the gap narrows to 50%.
• One twentieth of Sonnet 5.5: for prompts up to 100,000 tokens, Haiku 5.5's token price is 1/20 of Sonnet 5.5's.
• Same price class as GPT-6 Luna: both list at $0.15 per 1M input and $0.75 per 1M output; let quality on your own prompts decide.

Monthly example: a classification pipeline that processes 50 million input and 10 million output tokens a month, with prompts under 100,000 tokens, pays $15 on Haiku 5.5 (≈741.31 TL), $150 on Haiku 4.5 (≈7,413.06 TL) and $300 on Sonnet 5.5 (≈14,826.12 TL). This is a list-price comparison that assumes the same token count on all three models: according to Anthropic, Haiku 5.5 counts the same text as roughly 30% more tokens than Haiku 4.5, so the real gap is smaller than the table ratio. Use the cost calculator for your own volumes.

How the 100,000-Token Tier Affects Your Bill

In Anthropic's pricing the threshold looks at the size of the prompt: if the prompt exceeds 100,000 tokens, all input and output tokens of that request are billed at the higher tier. Onysoft applies the same rule: $0.15 / $0.75 per 1M tokens for prompts up to 100,000 tokens, $0.75 / $3.75 above that.

• Split long documents: splitting a 150,000-token document into parts under 100,000 tokens cuts the unit price to one fifth. Splitting can affect results, so compare the output on your own data.
• Watch conversation history: in long multi-turn chats the prompt grows every turn; summarizing history (context compaction) to stay under 100,000 tokens keeps you in the lower tier.
• Still economical above the threshold: even for prompts above 100,000 tokens, Haiku 5.5's price ($0.75 / $3.75) is a quarter of Sonnet 5.5's.

Every response returns its cost in the cost field; use it together with usage.prompt_tokens to see whether the higher tier applied.

Moving from Haiku 4.5 to Haiku 5.5

The only thing that has to change in your code is the model name: anthropic/claude-haiku-5.5 instead of anthropic/claude-haiku-4.5. Still, watch four things:

1. Thinking is on by default: according to Anthropic's documentation, adaptive thinking is on by default on Haiku 5.5, which was not the case on Haiku 4.5. Thinking tokens are billed as output, so output tokens per request can rise. Measure cost from the usage and cost fields in the response, not from list prices; if needed, try a lower effort through the reasoning parameter and compare.
2. Context grew from 200K to 1M, with a tier: long documents that did not fit Haiku 4.5 can now go in one request; remember that prompts above 100,000 tokens are billed at 5x.
3. Migrate gradually: anthropic/claude-haiku-4.5 stays active in the catalog, so you can compare on a sample, move traffic step by step and roll back with a one-line change.
4. The same text uses more tokens: according to Anthropic's documentation, Haiku 5.5 uses a newer tokenizer, so the same text counts as roughly 30% more tokens than on Haiku 4.5. Budget from the usage values in Haiku 5.5 responses rather than from Haiku 4.5 token counts; long prompts also reach the 100,000-token threshold sooner.

Alias: if you do not want to update code for every release, use the ~anthropic/claude-haiku-latest alias (the leading ~ is required). As of October 17, 2026 it is listed at the same price as Haiku 5.5. If you do not want behavior to change between versions on its own, calling the fixed ID is more predictable.

Where Haiku 5.5 Fits: Sub-Agents and High-Volume Work

Workload	Recommendation	Why
Classification, labeling, short summaries	anthropic/claude-haiku-5.5	The high-volume work Anthropic highlights for Haiku 5.5; the lowest unit price in the family.
Live customer support, speed-sensitive chat	anthropic/claude-haiku-5.5	The speed-sensitive work Anthropic recommends it for.
Sub-tasks in a coding agent such as reading files, search and summaries	anthropic/claude-haiku-5.5 (sub-agent)	The pairing Anthropic recommends when the main agent is Opus 5.5 or Sonnet 5.5.
Complex agentic coding, multi-step reasoning	anthropic/claude-sonnet-5.5 or anthropic/claude-opus-5.5	The better choice for this work, according to Anthropic.

Sub-agent cost scenario (list prices only): suppose a coding agent uses 6 million input and 1.2 million output tokens on one task. Running all of it on Opus 5.5 costs $72 (≈3,558.27 TL). If the planning and decision part — 1 million input and 200,000 output tokens — runs on Opus 5.5 and the reading and summarizing part — 5 million input and 1 million output tokens — runs on Haiku 5.5 (assuming Haiku prompts stay under 100,000 tokens), the cost drops to $13.50 (≈667.18 TL). This does not measure quality; validate the sub-agent's results on your own test set.

If you want the model chosen per request on mixed traffic, OnyRouter (onysoft/auto) is an option; for Opus 5.5 and Sonnet 5.5 details, see the Claude Opus 5.5 and Sonnet 5.5 API guide.

First Request with Python and curl

To get started, create an account, generate an API key with the sk-ony- prefix in the dashboard and top up a lira balance. Because the Onysoft endpoint follows the OpenAI schema, the official openai Python package works directly in streaming mode:

from openai import OpenAI

client = OpenAI(
    base_url="https://api.onysoft.com/v1",
    api_key="sk-ony-YOUR_KEY",
)

stream = client.chat.completions.create(
    model="anthropic/claude-haiku-5.5",
    messages=[
        {"role": "system", "content": "Classify the incoming support ticket as billing, technical or refund and write a one-line summary."},
        {"role": "user", "content": "The balance I topped up yesterday is not showing in my account, but my card was charged."}
    ],
    max_tokens=1024,
    stream=True,
)
for chunk in stream:
    if chunk.choices and chunk.choices[0].delta.content:
        print(chunk.choices[0].delta.content, end="")

Non-streaming requests: Onysoft returns non-streaming responses in a success/data envelope; the OpenAI-format response object sits in the data field. The example below sends the reasoning parameter through extra_body (forwarded unchanged to models that support it) and prints the prompt token count so you can watch the 100,000 threshold:

raw = client.chat.completions.with_raw_response.create(
    model="anthropic/claude-haiku-5.5",
    messages=[{"role": "user", "content": "Summarize the meeting notes below in five bullet points: ..."}],
    max_tokens=1024,
    extra_body={"reasoning": {"effort": "low"}},
)
data = raw.http_response.json()["data"]
print(data["choices"][0]["message"]["content"])
print("prompt tokens:", data["usage"]["prompt_tokens"])  # above 100,000 = higher tier

A non-streaming request to the same endpoint with curl. The content sits under the data field; for streaming, add "stream": true to the body:

curl https://api.onysoft.com/v1/chat/completions \
  -H "Authorization: Bearer sk-ony-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-haiku-5.5",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Label this product review as positive, negative or neutral."}]
  }'

See the API documentation for supported parameters and error codes, the model catalog for all 836 models available with the same key, and the Claude models page for the whole family.