Claude Haiku 5.5 in 2026: New Pricing, Copilot Access & Best Uses

Claude Haiku 5.5 is out, and it is Anthropic’s most aggressive pricing move yet. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output tokens. That is 90% below Haiku 4.5 and exactly matches OpenAI’s GPT-6 Luna, which launched last month. On launch day (October 7, 2026) it also went live in GitHub Copilot, so millions of developers can use it without an API key. If you run agents, subagents or any high-volume pipeline, re-check your model choices this week.

What’s actually new in Claude Haiku 5.5

Haiku 5.5 is the third model in the Anthropic Claude 5.5 models lineup. Opus 5.5 came first in September, and Sonnet 5.5 shipped on September 28. Anthropic promised that the smaller models would bring “many of the same improvements to performance, efficiency, and safety.” Haiku 5.5 delivers that promise at the cheap tier. It is available today on the Claude Platform as claude-haiku-5-5, and on AWS, Google Cloud and Microsoft Azure.

The capability jump over Haiku 4.5 is large. In published benchmarks, Haiku 5.5 scores 72.4% on OSWorld 2.1 computer use, up from 15.7% for Haiku 4.5. It scores 39.2% on Terminal-Bench 4.0, where Haiku 4.5 scored zero. On Humanity’s Last Exam with tools it scores 57.4%, up from 18.7%. It is also the first Haiku with an adjustable effort setting (low, medium, high, xhigh, max), so you decide per request whether to spend tokens on intelligence or save them. Its knowledge cutoff is June 2026, and it outputs text only.

Anthropic paired the launch with a second price cut. Cache reads on Sonnet 5.5 fell 50%, from $0.20 to $0.10 per million tokens, which Anthropic says lowers the cost of most agentic Sonnet 5.5 workloads by about 20%. The AI price war between Anthropic and OpenAI is now fought on per-token rates, not just benchmark charts.

Why it matters

• It’s the cheapest Claude model in 2026, by a lot. At $0.10/$0.50 per million tokens, Haiku 5.5 costs one-tenth of Haiku 4.5’s $1/$5 for normal-length prompts. Anthropic puts the average saving at about 75% once long-context requests are included.
• Watch the 100K cliff. Above 100,000 input tokens, every rate rises fivefold: $0.50 input, $2.50 output, $0.05 cache reads and $0.625 cache writes. That is still half the price of Haiku 4.5, but if your app routinely sends huge contexts, the savings shrink fast.
• It beats OpenAI on quality at the same price. GPT-6 Luna has the same list price. In Anthropic’s published numbers, Haiku 5.5 beats it on every benchmark where both have a score: OSWorld 2.1 (72.4% vs 48.9%), Terminal-Bench 4.0 (39.2% vs 16.4%) and GDPval-AA v2.1 (1620 vs 1437).
• Subagents just got almost free. GitHub says that in early testing Haiku 5.5 “matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps.” Fan-out tasks like search, file triage, test runs and summarisation no longer need a mid-tier model.
• Latency is a feature. Asana engineers reported 30%+ lower latency and up to 2.5x faster inference per agent turn. In interactive tools, a fast, decent answer often beats a slow, perfect one.
• Cheap caching changes prompt design. Cache reads cost $0.01 per million tokens. A long system prompt reused across thousands of calls now costs almost nothing, so put stable instructions up front and cache them.

How to use Claude Haiku 5.5 today

Call it from the API

Change the model string to claude-haiku-5-5. Nothing else in the request has to change.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-haiku-5-5",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "Classify this support ticket as billing, bug, or feature request: ..."}
    ]
  }'

Tune the effort setting

Effort is the main lever on Haiku 5.5. Use low effort for classification, extraction and routing. Use high or above for multi-step agent work. Here is the Python SDK version:

import anthropic

client = anthropic.Anthropic()

resp = client.messages.create(
    model="claude-haiku-5-5",
    max_tokens=2048,
    output_config={"effort": "low"},  # low | medium | high | xhigh | max
    messages=[{"role": "user", "content": "Extract name, email, and order ID as JSON: ..."}],
)
print(resp.content[0].text)

Check the current Anthropic docs for the exact parameter shape in your SDK version before you ship. Start at low and move up only when output quality fails.

Cache your system prompt

With cache reads at $0.01 per million tokens, this is the easiest saving available:

resp = client.messages.create(
    model="claude-haiku-5-5",
    max_tokens=1024,
    system=[{
        "type": "text",
        "text": LONG_STYLE_GUIDE_AND_RULES,
        "cache_control": {"type": "ephemeral"}
    }],
    messages=[{"role": "user", "content": user_input}],
)

Use it in GitHub Copilot

Claude Haiku 5.5 is generally available to Copilot Pro, Pro+, Max, Business and Enterprise users in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, GitHub.com, GitHub Mobile, the Copilot CLI and the cloud agent. In the IDE, open the Copilot Chat model picker and select Claude Haiku 5.5. In the Copilot CLI, switch models inside a session:

/model
# then pick Claude Haiku 5.5 from the list

It is billed at provider list pricing under usage-based billing. Business and Enterprise admins: new models are enabled by default unless you have turned default-enable off globally or disabled this model in your Copilot model policies. Rollout is gradual, so if you don’t see it yet, wait a day before filing a ticket.

Make it your Claude Code subagent model

Keep Opus or Sonnet as the planner and push the grunt work down to Haiku. Run a one-off session on it:

claude --model claude-haiku-5-5

Or pin a dedicated subagent in .claude/agents/test-runner.md:

---
name: test-runner
description: Runs the test suite, reads failures, and reports a short summary of what broke and where.
model: haiku
---
Run the project's tests. For each failure, report the file, the test name,
and a one-line likely cause. Do not attempt fixes.

Route by task, not by habit

A simple router prompt on Haiku can decide which model handles each request:

You are a router. Read the request and reply with exactly one word:
HAIKU  - extraction, classification, formatting, short Q&A, simple edits
SONNET - multi-file code changes, analysis needing judgment, long-form writing
OPUS   - architecture decisions, hard debugging, high-stakes reasoning

Request: {{user_request}}

How it compares: Haiku 5.5 vs Sonnet 5.5, Haiku 4.5 and GPT-6 Luna

Model	Input / Output (per 1M tokens)	OSWorld 2.1	Terminal-Bench 4.0	Best for
Claude Haiku 5.5	$0.10 / $0.50 (up to 100K); $0.50 / $2.50 above	72.4%	39.2%	Subagents, routing, extraction, quick edits, high-volume pipelines
Claude Sonnet 5.5	$2.00 / $10.00 (cache reads now $0.10)	83.9%	70.6%	Main coding agent, complex multi-step work
Claude Haiku 4.5	$1.00 / $5.00	15.7%	0.0%	Legacy; migrate off it
OpenAI GPT-6 Luna	$0.10 / $0.50	48.9%	16.4%	OpenAI-native stacks at the same price point

Our take: for routine coding, the Haiku 5.5 vs Sonnet 5 gap is now small enough that cost should decide. GitHub reports Haiku 5.5 matching Sonnet 5 on many coding tasks. On hard agentic benchmarks, though, Sonnet 5.5 stays well ahead (70.6% vs 39.2% on Terminal-Bench 4.0). Use Sonnet 5.5 where reasoning depth matters and Haiku 5.5 everywhere else. If you still run Haiku 4.5, switch now: it costs more and performs far worse.

What’s next

Haiku 5.5 completes the Claude 5.5 family. Expect fast follow-on work across the tooling: Claude Code defaults, Bedrock and Vertex quota increases, and third-party IDE integrations usually catch up within weeks. Watch for a -latest style alias update and for batch API pricing. At these list rates, batch discounts would make bulk classification and data-labelling jobs nearly free.

On the price war, OpenAI matched Anthropic at $0.10/$0.50 with Luna, and Anthropic answered with better benchmarks at the same price plus a Sonnet cache cut. The next round will likely be fought on long-context pricing. The 100K threshold is Haiku 5.5’s weak point, and a competitor removing that cliff would be the obvious counter. Expect pressure on the middle tier too: Sonnet 5.5 cache reads already dropped by half, and a base-rate cut could follow.

For builders, the practical move is to make your stack model-agnostic now. Put the model ID in config, log tokens per task, and re-run your own evals each time a cheaper model ships. The teams winning on margin in 2026 re-benchmark quarterly instead of trusting last year’s defaults.

Frequently Asked Questions

How much does Claude Haiku 5.5 cost?

For requests up to 100,000 tokens, Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Cache reads cost $0.01 and cache writes $0.125. Above 100,000 tokens, all rates rise fivefold: $0.50 input, $2.50 output, $0.05 cache reads and $0.625 cache writes.

Is Claude Haiku 5.5 available in GitHub Copilot?

Yes. It became generally available on October 7, 2026 for Copilot Pro, Pro+, Max, Business and Enterprise. It works in VS Code, Visual Studio, JetBrains, Xcode, Eclipse, GitHub.com, GitHub Mobile, the Copilot CLI and the cloud agent. It is billed at provider list pricing under usage-based billing, and rollout is gradual.

What’s the API model ID?

claude-haiku-5-5. It is available on the Claude Platform, AWS, Google Cloud and Microsoft Azure.

Should I use Haiku 5.5 or Sonnet 5.5?

Use Haiku 5.5 for high-volume, latency-sensitive or well-defined tasks: subagents, extraction, classification, quick code edits and terminal chores. Use Sonnet 5.5 as your main agent for complex, multi-step reasoning. It scores much higher on hard agentic benchmarks and costs 20 times more per token.

Is Haiku 5.5 the cheapest Claude model in 2026?

Yes. At $0.10/$0.50 per million tokens for standard-length prompts, it is the lowest-priced model in Anthropic’s current lineup and about 75% cheaper than Haiku 4.5 on average.

How does Haiku 5.5 compare to OpenAI’s GPT-6 Luna?

They share the same list price. In Anthropic’s published results, Haiku 5.5 beats Luna on all six benchmarks where both have scores, including OSWorld 2.1 (72.4% vs 48.9%) and Terminal-Bench 4.0 (39.2% vs 16.4%). Test on your own workload before switching.

Sources: GitHub Changelog, SiliconANGLE, PPC Land, Unite.AI, Anthropic.