Claude Haiku vs Sonnet: Speed, Cost and When to Use Each

Claude Haiku vs Sonnet compared: Haiku 5.5 and Sonnet 5.5 prices, the 100K-token Haiku tier, speed, default effort, and which tasks suit each model best.

Josh Pocock7 min read

Haiku vs Sonnet cover

AI Coding

Haiku vs Sonnet

Stride AI Academy

In Claude Haiku vs Sonnet, Haiku is the fast, cheap model and Sonnet is the stronger all-rounder. Claude Haiku 5.5 starts at $0.10 per million input tokens and $0.50 per million output tokens, against $2 and $10 for Claude Sonnet 5.5, so Haiku can be 20 times cheaper. Use Haiku for high-volume work like classification, extraction, routing and subagent tasks. Use Sonnet for everyday coding, data analysis, writing and agentic tool use.

This guide compares the two on price (including Haiku's 100K-token pricing tier), speed, thinking and effort, then shows where each fits, with worked cost examples and how Opus compares. Everything here was checked against Anthropic's models overview, Haiku 5.5 and Sonnet 5.5 pages, choosing a model, pricing, API errors and Claude's pricing page on November 22, 2026.

Claude Haiku vs Sonnet at a glance

Haiku 5.5 Sonnet 5.5
API model ID claude-haiku-5-5 claude-sonnet-5-5
Docs description "For high-volume, latency-sensitive tasks such as classification, extraction, and routing" "The best combination of speed and intelligence"
Input price $0.10 (prompts up to 100K tokens), $0.50 over $2
Output price $0.50 (prompts up to 100K tokens), $2.50 over $10
Comparative latency Fastest Fast
Default effort medium high
Thinking Adaptive, can be turned off Adaptive, lowest setting is between_tools
Context window 1M tokens 1M tokens
Max output 128K tokens 128K tokens
Knowledge cutoff June 2026 June 2026
Released October 7, 2026 September 28, 2026

Prices are US dollars per million tokens (MTok).

Claude Haiku API pricing and the 100K tier

Haiku 5.5 is the only current model priced by prompt length. From the pricing docs:

Haiku 5.5 Prompts up to 100,000 tokens Prompts over 100,000 tokens
Input $0.10 $0.50
Output $0.50 $2.50
5-min cache write $0.125 $0.625
1-hour cache write $0.20 $1
Cache hit $0.01 $0.05
Batch input / output $0.05 / $0.25 $0.25 / $1.25

Three rules from the docs decide which tier you pay:

  • Prompt length counts all input tokens, including cache reads and cache writes.
  • Each request is priced on its own. A request over 100K pays the higher prices even when part of its prompt is a cache hit.
  • Earlier requests keep their prices. Crossing the line on turn ten doesn't reprice turns one to nine.

So Haiku is 20 times cheaper than Sonnet on short prompts, and 4 times cheaper on long ones. If you feed Haiku big documents, keep prompts under 100K tokens where you can.

Sonnet 5.5 has one flat price: $2 input, $10 output, $2.50 and $4 for 5-minute and 1-hour cache writes, and $0.10 for cache hits. Batch is $1 and $5. For every other Sonnet version, see Claude Sonnet pricing.

Claude Haiku 4.5 API price

Haiku 4.5 is still listed as active on the deprecations page. It costs $1 input and $5 output per million tokens, with $0.10 cache hits and a 200K context window. Haiku 5.5 is cheaper on prompts up to 100K tokens and has a 1M context. The Haiku 5.5 page notes that it uses the newer tokenizer, so the same text counts as about 30% more tokens than on Haiku 4.5.

Speed, thinking and effort

The models overview rates Haiku 5.5 "Fastest" and Sonnet 5.5 "Fast" for comparative latency. Actual speed depends on prompt length, output length and effort.

Effort is where the two differ most in practice. Haiku 5.5 defaults to medium and Sonnet 5.5 to high. The choosing-a-model docs say to start at each model's default and adjust based on your evals, and that tuning effort is often a better lever than switching models.

Thinking works differently too, per the API errors docs:

  • Haiku 5.5 accepts thinking: {"type": "disabled"}, but not at xhigh or max effort.
  • Sonnet 5.5 rejects disabled. Its lowest setting is between_tools, which turns off up-front thinking and works at high effort or below.

The Haiku 5.5 page says to omit temperature, top_p and top_k, and the Sonnet 5.5 page says setting them returns a 400 error. Sonnet 5.5 is also on the docs' list of models that don't support forced tool use (tool_choice of any or tool). Haiku 5.5 isn't on that list.

When to use Haiku

The choosing-a-model matrix points to Haiku 5.5 when you need "the lowest latency and price." Its example uses:

  • Real-time applications
  • High-volume intelligent processing, such as extraction
  • Classification and routing, deciding which model or queue a request goes to
  • Subagent tasks, where a stronger model hands off bulk work
  • Cost-sensitive deployments that still need decent reasoning

The docs also describe an "efficiency-first" approach: start with Haiku 5.5, test thoroughly, and upgrade only for specific capability gaps. That suits prototyping, tight latency budgets and straightforward high-volume tasks.

When to use Sonnet

The same matrix points to Sonnet 5.5 for "speed and capability for everyday coding, agent, and enterprise workloads." Its examples:

  • Code generation
  • Data analysis
  • Content creation
  • Visual understanding
  • Agentic tool use

Sonnet is the safer pick when a task has several steps, needs judgement, or produces something a person will read closely.

Claude Haiku vs Sonnet vs Opus

Haiku 5.5 Sonnet 5.5 Opus 5.5
Input / output From $0.10 / $0.50 $2 / $10 $4 / $20
Latency Fastest Fast Moderate
Default effort medium high medium
Built for Classification, extraction, routing, subagents Everyday coding, agents, enterprise work Long-running agentic coding and knowledge work

The models overview says that if you're unsure, start with Opus 5.5 for most workloads. The pricing docs' own rule of thumb is Haiku for simple tasks, Sonnet for most production workloads and Opus for the most complex reasoning. For the top two, read Claude Sonnet vs Opus, and for the whole lineup, Claude models.

Use Haiku and Sonnet together

You don't have to pick one. The choosing-a-model docs describe an orchestrator pattern: a stronger model plans and delegates bulk work to lower-cost workers, so most tokens bill at the lower rate. Sonnet or Opus decides, Haiku does the repetitive part.

In Claude Code, the model configuration docs say the haiku alias points to Haiku 5.5 and sonnet to Sonnet 5.5 on the Anthropic API. You can keep a custom subagent on Haiku by setting model in its definition, while your main session stays on a stronger model. Our Claude Code subagents tutorial shows how.

Worked cost examples

Classifying 100,000 support tickets. Each ticket is about 1,000 input tokens with a 50-token label back. That's 100 million input and 5 million output tokens, every prompt well under 100K.

Input Output Total
Haiku 5.5 100 MTok × $0.10 = $10 5 MTok × $0.50 = $2.50 $12.50
Sonnet 5.5 100 MTok × $2 = $200 5 MTok × $10 = $50 $250

Run it through the Batch API and both totals halve.

One long document. A single 150,000-token prompt with a 2,000-token summary. Haiku pays its over-100K rate here.

Input Output Total
Haiku 5.5 0.15 MTok × $0.50 = $0.075 0.002 MTok × $2.50 = $0.005 $0.08
Sonnet 5.5 0.15 MTok × $2 = $0.30 0.002 MTok × $10 = $0.02 $0.32

The gap shrinks from 20x to 4x on long prompts. If quality matters more than a few cents per document, Sonnet's price is easy to justify there.

Haiku and Sonnet on Claude plans

On a subscription, you don't pay per token. Claude's pricing page lists both Haiku and Sonnet on every plan on November 22, 2026, including Free. Opus starts on Pro. See Claude pricing for the plans.

Learn to build with Claude Code

The Stride AI Coding course teaches Claude Code and Codex for building client tools, written for non-programmers, with weekly live calls to review your builds. Many of the free guides cover Claude Code setups, and the Claude Code community is where members compare theirs.

FAQ

What is the difference between Claude Haiku and Sonnet?

Haiku 5.5 is Anthropic's fastest and cheapest model, built for high-volume tasks like classification, extraction and routing. Sonnet 5.5 is slower but stronger, built for everyday coding, analysis and agent work. Both have a 1M-token context window and 128K max output.

How much cheaper is Haiku than Sonnet?

On prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 input and $0.50 output per million tokens against $2 and $10 for Sonnet 5.5, so 20 times cheaper. On prompts over 100,000 tokens, Haiku costs $0.50 and $2.50, so 4 times cheaper.

What is Claude Haiku API pricing?

Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 for longer prompts. Haiku 4.5 costs $1 and $5. The Batch API halves both.

Is Haiku good enough for coding?

For simple, well-defined pieces of a coding job, such as subagent tasks, it can be. Anthropic's docs recommend Sonnet 5.5 for code generation and agentic tool use and Opus 5.5 for complex agentic coding. Test Haiku on your own tasks before relying on it.

Should I use Haiku or Sonnet for subagents?

Anthropic's model selection guide lists subagent tasks as a Haiku 5.5 use case, with a stronger model as the orchestrator. Use Sonnet for subagents that need more judgement, such as code review or multi-step changes.

Can I use Haiku on the Claude Free plan?

Yes. Claude's pricing page lists Haiku and Sonnet on the Free plan, with limited usage. Opus needs Pro or higher.

Build it with us in Stride AI Academy

Four courses, weekly live calls and a community of people building and selling AI. Or start with 320+ free guides.

Keep reading

Free AI Coding guides