Skip to content
llmwise
Sign inStart free

Claude Haiku 5.5 · Writing

Claude Haiku 5.5 for writing: our test runs

Claude Haiku 5.5 ran our 5 writing prompts in llmwise, through the same pipeline your messages take. Here's every reply as it came, whether it passed, what it cost and what it counts as on Pro, next to the other models.

Start freeSee pricing

Based on 5 of our test runs on , through OpenRouter with the app's own prompt and settings. Updated .

Short answer

In our test runs on October 7, 2026, Claude Haiku 5.5 passed 3 of 5 of our writing prompts (1 of the 2 hard ones): tied 14th of 18 models. Each reply counted as 1 message on Pro, where it has 60 a day, and cost $0.0002 to run on average.

Prompt by prompt

Each of the 5 writing prompts, and how Claude Haiku 5.5's reply did.

Claude Haiku 5.5 on each writing prompt
PromptResultCostTime
Announce a second bakery shop on LinkedInEverydayPassed: Graded 4.5 of 5 on average (lowest 4).$0.00023.4 s
Rewrite corporate jargon in plain wordsEverydayFailed: Graded 3.3 of 5 on average (lowest 2).$0.00011.3 s
Decline a meeting and offer two timesEverydayPassed: Graded 5.0 of 5 on average (lowest 5).$0.00011.3 s
A product announcement with five rulesHardPassed: Graded 4.3 of 5 on average (lowest 4).$0.00011.5 s
Argue both sides of free busesHardFailed: Graded 4.0 of 5 on average (lowest 4); but a paragraph of 103 words, over the 90 allowed.$0.00033.3 s

Every writing reply from Claude Haiku 5.5

  1. Announce a second bakery shop on LinkedIn

    An everyday prompt: “Announce a second bakery shop on LinkedIn”, in full.

    Passed: Graded 4.5 of 5 on average (lowest 4).

    Big news from Crumb & Co.: we're opening a second shop in Leeds on 14 November!

When we started, we had one oven, a handful of loyal regulars, and a dream of making bread that people look forward to. Those regulars told their friends, and now we're ready to share the same warm loaves, slow-proved pastries, and morning buns with a whole new neighbourhood.

To celebrate, every customer on opening day gets a free coffee. Come in, say hello, and let us make you a cup.

Thank you to everyone who has supported us so far. We couldn't have done this without you, and we can't wait to see you in Leeds.

See you on the 14th.

    541 tokens in, 214 out · 3.4 s · $0.0002 · 1 message on Pro · answered by anthropic/claude-haiku-5.5 via Anthropic ·

  2. Rewrite corporate jargon in plain words

    An everyday prompt: “Rewrite corporate jargon in plain words”, in full.

    Failed: Graded 3.3 of 5 on average (lowest 2).

    Through teamwork across departments, our Q3 initiative shifted our approach to focus more on customers. Customer retention rose 12%, and customer loss slowed by 30 basis points across our enterprise segment. We also adjusted our multichannel customer contact points to use resources more efficiently.

    600 tokens in, 95 out · 1.3 s · $0.0001 · 1 message on Pro · answered by anthropic/claude-haiku-5.5 via Anthropic ·

  3. Decline a meeting and offer two times

    An everyday prompt: “Decline a meeting and offer two times”, in full.

    Passed: Graded 5.0 of 5 on average (lowest 5).

    Subject: Re: Vendor Demo on Wednesday

Hi Tom,

Thank you for the invitation to the vendor demo on Wednesday at 11:00. Unfortunately, I'm presenting to the board at that time, so I won't be able to attend.

Would Thursday at 10:00 or Friday at 14:00 work for you instead? I'd be glad to reschedule.

Thanks for understanding, and I look forward to hearing from you.

Best regards,
[Your Name]

    545 tokens in, 165 out · 1.3 s · $0.0001 · 1 message on Pro · answered by anthropic/claude-haiku-5.5 via Anthropic ·

  4. A product announcement with five rules

    A hard prompt: “A product announcement with five rules”, in full.

    Passed: Graded 4.3 of 5 on average (lowest 4).

    Introducing the Nordvik 2 desk lamp. It dims in five steps, so you can match the light to your work or your mood. Its arm turns a full 270 degrees, letting you aim light exactly where you need it. The lamp charges from USB-C, so one familiar cable keeps it ready to use. Priced at €79, the Nordvik 2 ships on 1 December. Could this be the desk lamp your workspace has been missing?

    557 tokens in, 143 out · 1.5 s · $0.0001 · 1 message on Pro · answered by anthropic/claude-haiku-5.5 via Anthropic ·

  5. Argue both sides of free buses

    A hard prompt: “Argue both sides of free buses”, in full.

    Failed: Graded 4.0 of 5 on average (lowest 4); but a paragraph of 103 words, over the 90 allowed.

    Free buses would remove the single biggest barrier to public transit: the fare itself. For low-income residents, fares consume a meaningful share of income, and free service means no one is forced to choose between a bus ride and groceries. Fare collection also slows boarding, and eliminating it speeds routes and improves reliability for everyone. Ridership tends to rise, which eases traffic, cuts emissions, and reduces the need for parking. Administrative costs of ticketing, enforcement, and fare evasion disappear too. Compared with the cost of building roads and subsidizing cars, funding buses directly is a cheap way to give every resident reliable mobility.

Free buses would divert public money from needs that matter more. Fare revenue, though often modest, funds service, and replacing it with taxes shifts costs onto everyone, including people who rarely ride. Free rides also tend to be used most by those who already have options, so the benefit may be less targeted than advocates claim. Without price signals, demand can outstrip capacity, leading to crowding and degraded service. Cities could instead offer fare discounts or subsidies for low-income riders, which target help precisely. Fixing frequency, safety, and maintenance would likely attract more riders than zero fares, and those dollars would stretch further.

    521 tokens in, 412 out · 3.3 s · $0.0003 · 1 message on Pro · answered by anthropic/claude-haiku-5.5 via Anthropic ·

How it compares on writing

Claude Haiku 5.5 beside the job's picks among every model. 3 models share the hard-problems pick, and Claude Haiku 5.5 isn't one of them.

Claude Haiku 5.5 beside other models on writing
ModelPassedOn ProCost per reply
DeepSeek V4 ProOur pick: hard problems (shared), best value5 of 5250 a month on Pro$0.0011
GPT-6 LunaOur pick: hard problems (shared), everyday5 of 560 a day on Pro$0.0001
Claude Haiku 5.53 of 560 a day on Pro$0.0002
On Pro: Pro's count on each model. Cost: what OpenRouter charged us per reply, on average.

Every limit is published. Paid plans also have a monthly fair-use limit on AI cost: Pro $7.50, Max $20, Ultra $42, Studio $85. Using every message on your plan at typical sizes stays under it; very large messages and heavy research use it faster. Every limit, explained.

Claude Haiku 5.5 next to each model on writing

Claude Haiku 5.5 against each model on the same 5 writing prompts. One run each, through OpenRouter: a wait depends on the provider and the load that day, so a lead under 10% counts as close.

  • Against Claude Fable 5.1: Claude Haiku 5.5 answered 3.9× sooner (1.5 s to 6.0 s), cost 107.2× less, and passed 3 of 5 to its 3. Only Claude Haiku 5.5 passed A product announcement with five rules. Only Claude Fable 5.1 passed Rewrite corporate jargon in plain words.

  • Against Claude Opus 5.5: Claude Haiku 5.5 answered 5.4× sooner (1.5 s to 8.2 s), cost 114.7× less, and passed 3 of 5 to its 4. Only Claude Opus 5.5 passed Rewrite corporate jargon in plain words.

  • Against Claude Sonnet 5.5: Claude Haiku 5.5 answered 1.5× sooner (1.5 s to 2.3 s), cost 21.2× less, and passed 3 of 5 to its 4. Only Claude Sonnet 5.5 passed Rewrite corporate jargon in plain words.

  • Against Claude Sonnet 5: Claude Haiku 5.5 answered 2.9× sooner (1.5 s to 4.5 s), cost 21.2× less, and passed 3 of 5 to its 4. Only Claude Haiku 5.5 passed Announce a second bakery shop on LinkedIn. Only Claude Sonnet 5 passed Rewrite corporate jargon in plain words; Argue both sides of free buses.

  • Against Claude Haiku 4.5: Claude Haiku 5.5 answered 1.5× sooner (1.5 s to 2.2 s), cost 7.1× less, and passed 3 of 5 to its 4. Only Claude Haiku 5.5 passed A product announcement with five rules. Only Claude Haiku 4.5 passed Rewrite corporate jargon in plain words; Argue both sides of free buses.

  • Against GPT-6 Astra: Claude Haiku 5.5 answered 3.1× sooner (1.5 s to 4.8 s), cost 69.1× less, and passed 3 of 5 to its 5. Only GPT-6 Astra passed Rewrite corporate jargon in plain words; Argue both sides of free buses.

  • Against GPT-6.1 Sol: Claude Haiku 5.5 answered 2.1× sooner (1.5 s to 3.3 s), cost 7.0× less, and passed 3 of 5 to its 4. Only GPT-6.1 Sol passed Argue both sides of free buses.

  • Against GPT-6 Sol: Claude Haiku 5.5 answered 2.3× sooner (1.5 s to 3.5 s), cost 14.1× less, and passed 3 of 5 to its 4. Only GPT-6 Sol passed Argue both sides of free buses.

  • Against GPT-6 Luna: Claude Haiku 5.5 answered 1.6× sooner (1.5 s to 2.4 s), cost 1.6× more, and passed 3 of 5 to its 5. Only GPT-6 Luna passed Rewrite corporate jargon in plain words; Argue both sides of free buses.

  • Against Gemini 3.1 Pro (preview): Claude Haiku 5.5 answered 6.0× sooner (1.5 s to 9.1 s), cost 57.0× less, and passed 3 of 5 to its 4. Only Gemini 3.1 Pro (preview) passed Rewrite corporate jargon in plain words.

  • Against Gemini 3.8 Flash: Claude Haiku 5.5 answered 2.5× sooner (1.5 s to 3.9 s), cost 8.2× less, and passed 3 of 5 to its 3. Only Claude Haiku 5.5 passed A product announcement with five rules. Only Gemini 3.8 Flash passed Argue both sides of free buses.

  • Against DeepSeek V4.1 Flash: Claude Haiku 5.5 answered 1.1× later (1.5 s to 1.3 s), cost 2.8× less, and passed 3 of 5 to its 4. Only DeepSeek V4.1 Flash passed Argue both sides of free buses.

  • Against DeepSeek V4 Pro: Claude Haiku 5.5 answered 2.6× sooner (1.5 s to 3.9 s), cost 6.7× less, and passed 3 of 5 to its 5. Only DeepSeek V4 Pro passed Rewrite corporate jargon in plain words; Argue both sides of free buses.

  • Against Grok 4.7: Claude Haiku 5.5 answered 3.9× sooner (1.5 s to 6.0 s), cost 35.7× less, and passed 3 of 5 to its 3.

  • Against Kimi K3: Claude Haiku 5.5 answered 1.9× sooner (1.5 s to 3.0 s), cost 23.8× less, and passed 3 of 5 to its 4. Only Claude Haiku 5.5 passed A product announcement with five rules. Only Kimi K3 passed Rewrite corporate jargon in plain words; Argue both sides of free buses.

  • Against GLM 5.3: Claude Haiku 5.5 answered 1.2× sooner (1.5 s to 1.8 s), cost 3.6× less, and passed 3 of 5 to its 4. Only GLM 5.3 passed Rewrite corporate jargon in plain words.

  • Against GLM 5.3 Flash: Claude Haiku 5.5 answered 5.5× sooner (1.5 s to 8.5 s), cost 2.0× less, and passed 3 of 5 to its 2. Only Claude Haiku 5.5 passed A product announcement with five rules.

How these runs were done

Rubric (graded). The grader model scores the reply from 1 to 5 on each published criterion. It passes with an average of 4 or more and no criterion under 3, and only if it also meets the prompt's automatic rules (length, words it must or mustn't use).

The grader is Claude Opus 5.5 at low reasoning effort; its own replies are graded by GPT-6 Astra, so no model grades itself. Its prompt and every rubric are on the methods page.

How the runs were done, and every writing prompt.

Claude Haiku 5.5, and writing, elsewhere

Questions

Is Claude Haiku 5.5 good for writing?

In our test runs it passed 3 of 5 writing prompts, tied 14th of the 18 models in llmwise. Every reply is on this page, so you can judge them yourself.

How many of my messages does a writing reply from Claude Haiku 5.5 use?

1 message each on Pro, where it has 60 a day on Pro. The price of a message is fixed and shown before you send it, however long the reply.

How were these runs done?

The same way for every model: each prompt sent through llmwise's own pipeline, each reply checked the same way. The methods page has every prompt and how each is scored.

Claude, GPT, Gemini, DeepSeek, Grok, Kimi, and GLM, in one chat.

See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.

Start freeSee pricing