Loading page…
The biggest discount applies to prompts up to 100,000 tokens. Higher token counts and adjustable reasoning make the rate card only part of the savings calculation.
Listen to this story
Haiku 5.5 is designed for frequent, bounded tasks and delegated work, not as Anthropic’s choice for the hardest coding jobs. Although its short-prompt rates are 90% below Haiku 4.5’s, Anthropic estimates average operating costs fall about 75%; a new tokenizer can count identical text as roughly 30% more tokens, so developers should remeasure workloads. The release also expands context and output limits, but migration requires API changes and capacity planning because Haiku 5.5 lacks Priority Tier.
Haiku 5.5 supports a one-million-token context window, up from 200,000, and up to 128,000 output tokens.
Anthropic reports 39.2% on Terminal-Bench 4.0 at maximum effort; VentureBeat notes its medium-effort score is roughly 20%, versus Sonnet 5.5’s 70.6%.
The model is available through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS.
Anthropic released Claude Haiku 5.5 on October 7, 2026, lowering the price of repeated AI work such as summaries, classification and smaller assignments inside larger projects. Its API rates start at $0.10 per million input tokens and $0.50 per million output tokens—a 90% cut from Haiku 4.5 for prompts up to 100,000 tokens.
The discount shrinks for longer prompts. Above 100,000 tokens, Haiku 5.5 charges $0.50 per million input tokens and $2.50 per million output tokens, half Haiku 4.5’s rates. Anthropic says roughly 90% of requests to the previous model fell into the shorter category.
Those rate cuts are not the same as savings on completed work. Anthropic estimates that Haiku 5.5 costs around 75% less to run on average, accounting for request sizes and changed token consumption. Tokens are the billable units into which the model divides content.
The model’s updated tokenizer counts the same input text as approximately 30% more tokens than Haiku 4.5, depending on the content. Anthropic’s documentation therefore tells developers to recount prompts and revisit output limits and cost estimates rather than reuse their old measurements.
Published input rates fall from $1.00 for Haiku 4.5 to $0.10 for Haiku 5.5.
Published output rates fall from $5.00 for Haiku 4.5 to $0.50 for Haiku 5.5.
Anthropic positions Haiku for narrowly scoped, high-volume jobs rather than the hardest coding assignments. Financial AI company Rogo supplied one example: a larger model builds a presentation while a Haiku subagent—a model assigned one piece of the work—retrieves the revenue figure needed for a slide.
Haiku 5.5 can hold one million tokens of context, up from 200,000 in Haiku 4.5, and produce up to 128,000 output tokens, twice its predecessor’s limit. It also introduces adjustable effort to the Haiku family. Adaptive thinking is enabled by default, with medium effort as the starting setting.
The effort setting matters when reading performance claims. Anthropic’s table gives Haiku 5.5 a 39.2% score on Terminal-Bench 4.0, a test of multistep command-line work, against Sonnet 5.5’s 70.6%. VentureBeat notes that Haiku’s chart places the roughly 39% result at maximum effort, while medium scores roughly 20%. These are vendor-reported results.
Anthropic still recommends Sonnet and Opus for complex coding tasks. Its speed claim has a separate boundary: Haiku 5.5 is its fastest model at standard speeds, but Opus models running in Fast Mode are faster.
Haiku 5.5 is available through the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS. Its Claude API identifier is claude-haiku-5-5. For existing integrations, the migration guide details changes beyond replacing the model name:
There is also a capacity-planning distinction: Haiku 5.5 does not support Priority Tier. Organizations with a Haiku 4.5 commitment under that service must plan capacity separately.
The coordinated announcement also halves Sonnet 5.5 cache-read rates to $0.10 per million tokens. Cache reads reuse stored context; Anthropic estimates around 20% savings on most agent workloads. Monthly API credits roll out this week: $100 for Max 5x, $200 for Max 20x and up to $500 pooled across Team users, usable on any model.
Story updates
haiku 5.5 failed bridgemind's lava lamp test. billions in compute, years of research, and they still can't solve the fucking lava lamp. what are we even doing anymore. i'm calling for a complete pause on anthropic's training runs until we figure out what went wrong here.
Claude Max and Team subscribers are getting a monthly budget to build their own AI apps and agents, separate from their chat allowance. Anthropic announced the API-credit rollout on October 7, 2026, with Team balances reaching $500 a month. The benefit funds developer use—not more chat capacity or interactive Claude Code sessions.
The credits pay for calls to the Claude API, the interface that lets software use Claude. They also cover Claude Managed Agents, the Claude Agent SDK and the Console playground. Anthropic says they work with any available Claude model and are intended to support experiments with tools, applications and agents.
The boundary is where that work runs. Credits apply on Anthropic’s own Claude Platform, not Amazon Bedrock, Google Cloud or Microsoft Foundry. They do not cover interactive Claude Code sessions or extra usage in the Claude apps, and they leave existing subscription usage limits unchanged.
Team balances combine $20 per Standard seat and $100 per Premium seat, including discounted Nonprofit and Scientists plans. Three Standard seats and two Premium seats produce $260 a month. Seats added during a billing month increase the next month’s credit, rather than the current balance.
Anthropic says access is rolling out over a few days. An eligible subscription must be active and in good standing; new subscribers must wait seven days. Web, iOS and Android subscriptions qualify, but claiming happens in a web browser. Free, Pro and Enterprise plans are excluded.
The Help Center instructions require the Max subscriber or a Team Owner or Primary Owner to claim. That person also needs an Owner, Admin or Billing role in the receiving Console organization. No payment method is required on Claude Platform.
Open claude.ai billing settings: Settings > Billing for Max, or Organization settings > Billing for Team.
In API credits, select Link organization. Choose an existing Console organization or create one, then accept the credit terms.
Find the credits in the Console’s billing balance. Create an API key to start making requests.
Each plan can fund only one Console organization, and each organization can receive credits from only one plan. Choose carefully: changing the linked organization requires support. Every API key and workspace in that organization draws from the same balance; workspace spending limits can restrict individual projects.
Credits arrive shortly after each subscription payment, or monthly on annual plans. Unused amounts expire at the end of the credit cycle, without rollover. Included credits are spent before purchased credits. Auto-reload still follows the purchased balance, so it can add funds even while included credits remain.
Anthropic’s platform documentation says credited usage still counts toward the organization’s monthly spending cap. Those caps reset on the first of the calendar month, while credits follow the billing cycle. Linked organizations move to at least the Start usage tier, or Build when monthly credits exceed $200; the credits do not otherwise earn higher tiers.
When the balance runs out, requests continue against purchased credits or stop if none remain. API usage is never charged to the Claude subscription. Organizations invoiced through Anthropic sales instead have usage beyond the credits billed as usual.
anthropic.comHaiku 5.5 notes - ~75% cheaper than Haiku 4.5 - 1M context (up from 200K) - Computer use: 15.7% to 72.4% - First Haiku with an effort setting (low to max) Good uses: 1. Subagents under Opus. Opus plans, and Haiku agents run the research and drafts in parallel. Why: cheap and fast enough to run many at once. 2. Browser tasks. Flight searches, form fills, clicking through web apps. Why: computer use jumped from 15.7% to 72.4%. 3. Reservations and errands. Find a restaurant, check the weather, book the table. Why: it can now work real websites 4. Long transcripts. A full day of calls goes in, and the action items come out. Why: 1M context fits it all in one pass. 5. Inbox sorting. 500 emails...
Loading discussion...
Join the conversation
Explain which would help you make a fairer comparison.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.