From Conversation to Control: How Anthropic Is Advancing AI Through Claude

Anthropic’s Claude lineup has moved on from Claude 3.5 Haiku, but the older model still matters to anyone comparing lightweight AI options. Anthropic Claude 3.5 Haiku was an earlier model for quick tasks, text work, and other knowledge work where speed mattered.

Its role can help you understand what lightweight Claude models were built to do. A new project, though, calls for a current model choice that fits its tasks, use cases, and response-time needs. The difference matters when you plan an app, compare results, or choose a model for daily work.

This overview explains the earlier model’s place and the newer Haiku options for new work. A useful decision starts by separating what the earlier model was from what is currently available.

What Is Anthropic Claude 3.5 Haiku?

Claude 3.5 Haiku was Anthropic’s lightweight Claude model, built for quick responses and practical work. Haiku is the speed-oriented tier in Anthropic’s model family, with a focus on useful intelligence and fast text responses.

Claude 3.5 Haiku was one model version, not the entire Claude service. Claude is Anthropic’s family of AI assistants and models, with Haiku, Sonnet, and Opus as distinct tiers.

A basic Claude 101 starts by separating the assistant from the models behind it. Claude is the assistant and model family; Claude 3.5 Haiku names one specific model in that family.

The word “Haiku” signals a tier, while “3.5” identifies a generation. Other versions can have different abilities, names, and access options, even when they share the Haiku label.

That distinction helps when you read product pages or model IDs. A tool may offer Claude through one interface, while a developer selects a specific model for an app.

The name identifies one generation in a family, not a promise that every version has the same features or availability.

Is Claude 3.5 Haiku Still Available?

Claude 3.5 Haiku is no longer available through the Claude API, where it was retired on February 19, 2026. That status applies to Anthropic’s API, not automatically to every third-party platform.

Anthropic introduced Claude 3.5 Haiku on October 22, 2024, and retired it from the Claude API on February 19, 2026. Anthropic’s policy gives at least 60 days’ notice before retiring public models on its own platforms.

That notice rule covers platforms run by Anthropic. A cloud provider or other service may have its own model listings and schedule, so API status alone does not settle availability everywhere.

Claude Haiku 4.5 followed as a newer Haiku generation. It launched on the Claude Platform on October 15, 2025, and Amazon Bedrock lists October 16, 2025, as its launch date.

Those dates refer to different platforms, rather than a single shared launch. A model may also remain listed on another provider after its Claude API access ends.

For an existing integration, confirm the status of the specific provider and endpoint before planning a migration.

What Does Claude Haiku 4.5 Do Well?

Claude Haiku 4.5 scored 73.3% on SWE-bench Verified, while Claude 3.5 Haiku scored 40.6% on the same named benchmark. Those results point to coding ability, but they are not a controlled head-to-head test.

Anthropic says Haiku 4.5 matched Sonnet 4 on coding, computer-use, and agent tasks. A benchmark frames performance on a set test; it cannot promise the same results in your app.

Coding and agent tasks

SWE-bench Verified tests how well a model handles real software issues. The named results suggest a notable difference between the two Haiku versions, but test setup and task fit still matter.

Tool use also changes what a model can do. In agentic work, a model can call tools, inspect results, and take steps toward a goal, rather than only returning text.

Where a fast Haiku model fits

Claude Haiku 5.5 lists use cases suited to quick responses and repeated tasks. Consider these examples:

• Classifying incoming text and summarizing documents.
• Routing requests to the right team, tool, or model.
• Running real-time chat, voice, or live support.
• Handling subagent work inside a larger process.
• Doing repetitive computer use or focused coding.

These tasks often benefit from speed and steady output across many prompts. For example, a support team might route routine questions before a person handles complex cases.

A fast model can also take a narrow coding task or operate a computer through repeated steps. More open-ended work may need stronger reasoning or more careful review.

Test the model on representative prompts and tool flows before relying on benchmark performance in production.

How Much Does Claude Haiku 4.5 Cost?

Claude Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens on the Claude API. Input and output have separate rates, so the amount of text the model returns affects cost.

A token is a small chunk of text that the API counts as input or output when calculating usage. The final bill reflects how many tokens your prompts and model responses use.

Haiku 4.5 API rates

For every million input tokens, the Haiku 4.5 API rate is $1. For every million output tokens, the rate is $5.

Repeated prompts can use prompt caching, which stores input for later requests. Haiku 4.5 cache writes cost $1.25 per million tokens for a five-minute cache or $2 for a one-hour cache.

Cache reads cost $0.10 per million tokens. The write rate depends on how long the cache lasts, while the read rate applies when the stored input is reused.

Caching, batch requests, and newer Haiku rates

Batch requests for Haiku 4.5 receive a 50% discount on input and output pricing. This option suits work that can run in groups instead of needing an immediate response.

Claude Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. For longer prompts, above 100,000 tokens, its rates are $0.50 per million input tokens and $2.50 per million output tokens.

Sonnet 5.5 costs $2 per million input tokens and $10 per million output tokens. Opus 5.5 costs $4 per million input tokens and $20 per million output tokens.

Among those three tiers, Haiku 5.5 has the lowest listed per-token rates. Your bill still depends on token volume, prompt length, caching, and batch use.

Estimate input and output separately, since output tokens carry a different rate from input tokens.

Is Haiku or Sonnet Better for Your Work?

Haiku is the practical choice when a task is frequent and latency-sensitive, while Sonnet or Opus fits work that needs more depth. The best fit depends on how much reasoning the work needs and how often you run it.

Haiku 5.5 is positioned as the fastest Claude 5.5 model for high-volume, latency-sensitive work. Its stated strengths include classification, extraction, and routing.

Use the task itself to guide the choice:

• Choose Haiku 5.5 for high-volume tasks, quick replies, and short or routine writing.
• Choose Sonnet 5.5 for well-scoped work that needs a balance of speed and intelligence.
• Choose Opus 5.5 for complex coding, extended reasoning, and demanding knowledge work.

Sonnet can suit writing that needs clear structure and careful development. Opus can be a better fit when writing also calls for deeper research or analysis.

Haiku can handle short drafts, summaries, and routine text when speed and volume matter more than extended reasoning. For harder work, a larger Claude model may produce a stronger result.

These roles are starting points, not fixed rules for every prompt. A short coding task may suit Haiku, while a long project plan may call for more analysis.

Choose by the difficulty and volume of the task, then evaluate the same representative work in each tier.

How Can You Access Claude Haiku?

Claude Haiku 5.5 is selectable on Claude’s web, iOS, and Android interfaces for users on Free, Pro, Max, Team, and Enterprise plans. Free access means using Haiku through an eligible Claude plan, not free API use.

The Free plan costs $0, while Pro costs $17 per month with annual billing or $20 with monthly billing. Paid plans have different prices and terms.

Using Claude on the web and mobile

Claude Haiku 5.5 is available through Claude’s web app and mobile apps for iOS and Android. The pricing page lists Haiku on Free, Pro, Max, and Team plans, while Enterprise users can also select it.

Max costs $100-$200 per month and includes $100-$200 per month in API credits. Team standard seats cost $20 per seat monthly with annual billing or $25 monthly with monthly billing.

Team premium seats cost $100 per seat monthly with annual billing or $125 monthly with monthly billing. Enterprise pricing is not included in these listed plan rates.

Using an API or cloud platform

Developers can reach Haiku through these routes:

• Claude API, with a model ID selected for the application.
• Amazon Bedrock global inference profile, model ID global.anthropic.claude-haiku-4-5-20251001-v1:0.
• Amazon Bedrock Mantle endpoint, model ID anthropic.claude-haiku-4-5.

Amazon Bedrock bills the model through AWS Marketplace. Charges appear on the AWS bill under the model provider.

So, Claude Haiku 4.5 is not automatically free because Haiku 5.5 can be selected in a Free plan. Free access applies to the consumer plan’s model choice, not a promise of free Haiku 4.5 API use.

Check the plan and endpoint you intend to use, since consumer access and developer billing are separate.

What Should Developers Check Before Deploying Haiku?

Claude Haiku 4.5 supports a 200K-token context window and up to 64K output tokens. That capacity sets the text it can take in and the maximum response it can return.

Claude Haiku 4.5 accepts text and image input, then produces text output. It supports manual extended thinking, but not adaptive thinking.

Its reliable knowledge cutoff is February 2025, while its training-data cutoff is July 2025. The two dates describe different limits, so they are not interchangeable.

Claude Haiku 4.5 is marked Active (legacy) in Claude Platform specifications, with retirement not scheduled sooner than October 15, 2026. That label does not prove an endpoint is currently accessible.

Its Claude API alias resolves to a pinned model snapshot. A developer-facing model identifier, or model ID, names the endpoint an application calls.

Amazon Bedrock lists reasoning support for Haiku 4.5. Claude Haiku 5.5 is the newer announced generation, announced on October 7, 2026.

Before building a new integration, check the exact model ID and the current endpoint status on your chosen platform. An alias and a displayed status label may not tell you whether a specific route will work.

Pin and test the exact model version used by an application rather than assuming an alias will always point to the same snapshot.

Conclusion

Claude 3.5 Haiku explains an earlier stage of Anthropic’s lightweight-model lineup, while a current project needs a current model choice. For new work, start with the current Haiku option and weigh speed, volume, and reasoning needs.

A model that handles many routine tasks quickly may not be the best fit for complex coding or deeper knowledge work. The right choice should match the real demands of your prompts and tools.

Run a small set of representative prompts through the intended model and endpoint before you commit. Test the workload you actually need to run, then select the Haiku, Sonnet, or Opus tier that meets its quality and latency needs.