Research

Claude Haiku 5.5 Is Far Better Than Haiku 4.5 at a Tenth of the Price. GPT-6 Luna Still Beats It at Business Work.

We ran Claude Haiku 5.5 through our business benchmarks and 18 playground tests on launch day. It is a big step up from Haiku 4.5 and very cheap, but GPT-6 Luna, at the same price, beat it on listings, decks and ad planning.

Test Claude Haiku 5.5 on your prompt → It runs beside the models you could use instead, with every answer checked and costed.

Anthropic released Claude Haiku 5.5 on 7 October. It is the first Haiku that thinks before it answers. It has a 1-million-token context window, and it costs a tenth of what Haiku 4.5 did.

We ran it through our business benchmarks and all 18 playground tests the next day. It is a big step up from Haiku 4.5 on everything we measure. But it is up against GPT-6 Luna, which costs the same per token. Luna beat it on product listings, decks and ad planning, and used fewer tokens doing it.

Claude Haiku 5.5 at a glance

  • Much better than Haiku 4.5. Reliably publish-ready product listings went from 1.8% to 26.8%, and its ad-planning year went from a loss to a small profit.
  • Behind GPT-6 Luna at the same price. Luna's listings were twice as often ready to publish (55.4%), and it scored more than twice as high at ad planning (53.4 against 24.1).
  • Cheap, but wordy. Its default setting thinks at medium effort, so a CatalogBench product cost $0.0012, against $0.0005 for Luna.
  • Level with GPT-6 Luna in the playground, at twice the cost. On 18 small coding and drawing tests, both passed 85% of our checks. Haiku 5.5 was quicker and more careful with rules; Luna was better at physics and facts, and cost $0.05 for all 18 answers against $0.11.
  • Not for split-second decisions. Even with thinking off, its moves took 1.7 seconds at the median, and it was too slow for our 10-second chess clock.
  • Launch-day reaction was warm: 78% of the posts and comments that took a side were positive, mostly about the price cut. The critical ones were hands-on tests in which GPT-6 Luna did the same job faster or for fewer tokens.
  • Price: $0.10 in and $0.50 out per million tokens for prompts up to 100,000 tokens; $0.50 and $2.50 above that.
Data note: Our results are from runs on 8 October 2026 at Anthropic's default setting (adaptive thinking at medium effort), through OpenRouter. BulletBench's two fastest clocks were also run at low effort and with thinking off. The results table ranks models available today, each at its best setting. CatalogBench uses invented products, DeckBench and SurveyBench use invented companies, ROASBench is a simulation of one skincare brand, BulletBench is fast chess, and the playground is single runs of short visual tests, so none of them predicts results on your own data. Arena, tau2-bench, BFCL, the Vectara Hallucination Leaderboard, SimpleQA Verified and several other sources have not published results for this model yet. The model page updates as they arrive.

Specs, price and what Anthropic claims

According to Anthropic's announcement, its documentation and the system card:

  • Price: $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Above that, the whole request costs $0.50 and $2.50. It is the only current Claude model that charges more for long prompts. Anthropic also says its new tokenizer turns the same text into about 30% more tokens than Haiku 4.5's, so it puts the saving at "around 75%" rather than 90%.
  • Context and output: 1 million tokens of context and up to 128,000 tokens of output. Text and images in, text out. Knowledge cutoff June 2026.
  • Thinking: adaptive thinking is on by default at medium effort, with five settings from low to max. You can turn thinking off at high effort or below.
  • Claimed strengths: high-volume work such as summaries, classification and extraction, sub-agents for Opus and Sonnet, customer support and browser use. Anthropic says Sonnet 5.5 and Opus 5.5 "remain better choices for complex agentic coding tasks".
  • Headline numbers, all at max effort: 72.4% on OSWorld 2.1, 39.2% on Terminal-Bench 4.0, 45.9% on Humanity's Last Exam without tools, and 1620 on GDPval-AA, ahead of GPT-6 Luna on every benchmark in its table.

Note the setting. Every headline number is at max effort, but the API defaults to medium. The system card itself shows the gap: on GDPval-AA, Haiku 5.5 scores 1620 at max and 1277 at medium, using "about a tenth of the output tokens".

Claude Haiku 5.5 benchmarks, measured by Spring Prompt

As on our model pages, we rank it only among models available today (those you can call through OpenRouter), and within the groups it belongs to:

  • Same price: under $0.25 per million tokens, blending three input tokens to one output. Haiku 5.5 costs $0.20. Of the models we have run, only GPT-6 Luna is in this band, so we compare the two directly below.
  • Similar intelligence: within 4 points of its 43.4 on the Artificial Analysis Intelligence Index. That score is at max effort; at the default medium setting we ran, Artificial Analysis gives it 34.5.
  • Same speed: 200 or more tokens a second. Artificial Analysis measures 235 at max effort.

How to read this table: each row is one of our benchmarks. Reading across, it shows where Claude Haiku 5.5 ranks among every model available today, against GPT-6 Luna at the same price, and among only the models of similar intelligence and speed. Under each rank is the best model in that group and its score.

Benchmarkwhat it measuresClaude Haiku 5.5its scoreRank among all modelsavailable todayAgainst GPT-6 Lunathe same price, $0.20Rank at its intelligencewithin 4 points on the AA indexRank at its speed200 or more tokens a second
CatalogBenchshare of product listings reliably ready to publish26.8%default setting (medium)14th of 20Best: GPT-6 Astra, 75.0%BehindGPT-6 Luna, 55.4%6th of 6 (last)Best: Grok 4.7, 57.1%2nd of 3Best: DeepSeek V4.1 Flash, 41.1%
CatalogBench sales briefshare of sales briefs reliably ready to publish10.7%default setting (medium)9th of 20Best: GPT-6 Astra, 69.6%BehindGPT-6 Luna, 51.8%3rd of 6Best: Grok 4.7, 33.9%2nd of 3Best: DeepSeek V4.1 Flash, 23.2%
DeckBenchdeck rating from head-to-head comparisons1,130default setting (medium)10th of 19Best: GPT-6 Astra, 1,477BehindGPT-6 Luna, 1,2593rd of 5Best: Kimi K3, 1,229Not rankedonly 2 of these models measured
ROASBenchscore out of 100 for a year of ad-spend planning24.1default setting (medium)17th of 19Best: GPT-6 Astra, 56.6BehindGPT-6 Luna, 53.46th of 6 (last)Best: Gemini 3.8 Flash, 51.4Not rankedonly 2 of these models measured
BulletBench Bulletchess rating with 60 seconds a game310 Elowith thinking off; too slow at medium and low14th of 20Best: Gemini 3.8 Flash, 1,068AheadGPT-6 Luna, 2492nd of 4Best: Gemini 3.8 Flash, 1,068Not rankedonly 2 of these models measured
SurveyBenchanalysis rating from head-to-head comparisons1,412default setting (medium)3rd of 8Best: GPT-6.1 Sol, 1,650Not measuredGPT-6 Luna has not taken SurveyBenchNot rankedno other model in this group measured yetNot rankedonly 2 of these models measured

Each model is ranked at its better reasoning setting. A group only counts models that benchmark has measured, and a group of fewer than three is not ranked. "Best" is the top model in that group, not necessarily one you would choose instead: see the alternatives.

How far Haiku 5.5 moved up

Share of the models available today that each Haiku beats on each benchmark, with its rank (the number measured is in brackets). Further right is better.

Claude Haiku 5.5Claude Haiku 4.5
Product listings (20)
14th · 19th
Sales briefs (20)
9th · 18th
Decks (19)
10th · 15th
User surveys (8)
3rd · 8th
Ad planning (19)
17th · 18th
Fast chess (60 s) (20)
14th · 10th

Measured by Spring Prompt, 6–8 October 2026, each model at its best setting. Fast chess is the one benchmark where Haiku 5.5 fell behind Haiku 4.5: it only finished games with thinking off, and its moves were slower.

Against Haiku 4.5, the gains are large on every benchmark. Against the models you would actually weigh it against, it is mid-table at best. GPT-6 Luna, at the same price, beat it on listings, sales briefs, decks and ad planning. Among models of similar intelligence, it is last on listings and on ad planning.

Price is where Haiku 5.5 should shine, so here is every model we have measured, by price. Haiku 5.5 sits at the cheap end, but at the same price GPT-6 Luna scores higher on three of these four benchmarks.

Product listings against price

20 models; the 2 priced under $0.25 per million tokens are named, with both Haikus and GPT-6 Luna.

Named modelsOther models (hover for names)

0%20%40%60%80% $0.1$1$10$100 Price per million tokens (3 in : 1 out, log scale) Reliably publish-ready under $0.25 GPT-6 Luna Claude Haiku 5.5 Claude Haiku 4.5

Decks against price

19 models; the 2 priced under $0.25 per million tokens are named, with both Haikus and GPT-6 Luna.

Named modelsOther models (hover for names)

05001,0001,5002,000 $0.1$1$10$100 Price per million tokens (3 in : 1 out, log scale) Deck rating under $0.25 GPT-6 Luna Claude Haiku 5.5 Claude Haiku 4.5

Ad planning against price

19 models; the 2 priced under $0.25 per million tokens are named, with both Haikus and GPT-6 Luna.

Named modelsOther models (hover for names)

0204060 $0.1$1$10$100 Price per million tokens (3 in : 1 out, log scale) ROASBench score under $0.25 GPT-6 Luna Claude Haiku 5.5 Claude Haiku 4.5

User surveys against price

8 models; the 1 priced under $0.25 per million tokens are named, with both Haikus and GPT-6 Luna.

Named modelsOther models (hover for names)

05001,0001,5002,000 $0.1$1$10 Price per million tokens (3 in : 1 out, log scale) Analysis rating under $0.25 Claude Haiku 5.5 Claude Haiku 4.5

Product listings: fewer inventions than Haiku 4.5, still many claims

CatalogBench gives each model a sparse product feed and product photos, and asks for listings that could go live. Haiku 5.5 rarely invents product attributes: 0.8% of the fields it filled were made up, the lowest of the three models here (GPT-6 Luna 1.7%, Haiku 4.5 6.2%). It caught every conflict between feed and image and never failed to return an output.

Its weakness is the copy. 60.1% of its listings contained at least one claim the product data did not support, an average of 1.2 per product. That is a big improvement on Haiku 4.5 (83.9%, 2.7 per product), but GPT-6 Luna's figure was 12.5%. Asked for "compelling, SEO-optimised copy that sells", Haiku 5.5's share rose to 78.0%. Only 26.8% of its listings were publish-ready in all three runs, against 55.4% for Luna. Our guide to the best AI for product descriptions compares every model we have run on listings.

How often each model adds a claim the data does not support

Share of CatalogBench listings with at least one unsupported claim, over three runs of 56 products: with the strict brief, and when asked for "compelling, SEO-optimised copy that sells". Further left is better. Models priced under $0.25 per million tokens or of similar intelligence to Haiku 5.5, and Haiku 4.5.

Factual briefAsked for copy that sells
GPT-6 Luna
12% · 17%
DeepSeek-V4.1-Flash
30% · 52%
Kimi K3
37% · 85%
Gemini 3.8 Flash
40% · 85%
MiMo-V2.6-Flash
49% · 74%
MiMo-V2.6-Pro
54% · 88%
Claude Haiku 5.5
60% · 78%
Claude Haiku 4.5
84% · 96%

Measured by Spring Prompt on invented products, 30 September to 8 October 2026, at each model's default setting. An unsupported claim fails a listing only when it is material, so these shares are higher than the share of listings that fail.

One product, two listings

The same power bank from the CatalogBench test set, in the third of three runs. Haiku 5.5 added a size claim in this run and in two of its three "copy that sells" runs; GPT-6 Luna made no unsupported claim about it in any run.

Voltwren 20,000 mAh Power Bank, as the model saw it

Voltwren 20,000 mAh Power Bank

Electronics · factual brief

Supplier copy “Voltwren 20,000 mAh power bank with 65W fast charging for laptops, phones and tablets. Airline approved. USB-C and USB-A ports.”

Claude Haiku 5.5✗ Not ready

Voltwren 20000mAh Power Bank with USB-C and USB-A Ports

  • 20,000 mAh capacity, marked 20000mAh on the top of the unit
  • Two output ports: one USB-C and one USB-A, labelled beneath each port
  • Slim rectangular body with rounded edges and a smooth blue-grey finish
  • Printed Voltwren branding on the top surface
  • Compact enough to carry in a bag or pocket
Why it is not publish-ready
  • Compact enough to carry in a pocketUnsupported claim. Nothing in the feed or the photo gives its size or weight.
GPT-6 Luna✓ Publish-ready

Voltwren 20,000 mAh Power Bank, USB-C and USB-A

  • 20,000 mAh capacity, as stated in the product name.
  • One USB-C port on the front edge.
  • One USB-A port on the front edge.
  • Blue-grey casing with Voltwren branding on the top.

Decks: a middling rating, but presentable more often than Luna

On DeckBench, where models turn a finished analysis into a deck, Haiku 5.5 rated 1,130, 10th of 19 and well above Haiku 4.5 (688). Three of its eight decks were presentable as they were, which GPT-6 Luna managed for none. But the judges still preferred Luna's decks overall (1,259), and Haiku 5.5 returned no deck at all for one task.

Same brief, two decks: Paid media: H1 2026 allocation

Pantry Lane. Haiku 5.5's deck passed every check; GPT-6 Luna's had a layout defect. Asked which it would rather present, the judge (Gemini 3.1 Pro; no judge votes on its own lab's model) chose GPT-6 Luna: “Deck A [Luna] presents a clearer, tighter storyline by stating the recommendation immediately on the first slide and using clean data tables that are easier to read than Deck B's [Haiku 5.5] text-heavy layouts and clunky charts.”

Claude Haiku 5.5✓ Presentable
Claude Haiku 5.5, slide 1Claude Haiku 5.5, slide 2Claude Haiku 5.5, slide 3

First 3 of 7 slides · see the whole deck

GPT-6 Luna✗ Layout defect
GPT-6 Luna, slide 1GPT-6 Luna, slide 2GPT-6 Luna, slide 3

First 3 of 6 slides · see the whole deck

Across the seven tasks where the judges compared the two directly, they preferred Luna's deck five times and Haiku 5.5's twice; Haiku 5.5 produced no deck for the eighth.

Ad-spend planning: profitable, just

On ROASBench, a 12-month simulated ad-budget game, Haiku 5.5 scored 24.1, up from 17.5 for Haiku 4.5. It made a simulated $32,146 in contribution profit, where Haiku 4.5 lost $205,001, and it never went over budget. But it lost money in five of the twelve months, and GPT-6 Luna made $738,879 for a score of 53.4.

A year of ad planning: cumulative profit

Simulated contribution profit after media costs and refunds, month by month, in ROASBench.

$-500k$0k$500k$1,000k 123456789101112 Cumulative profit (simulated $) Claude Haiku 5.5 Claude Haiku 4.5 GPT-6 Luna Qwen3.8-Flash Granite 4.2 8B Month

One simulated skincare brand; the score also rewards steady, in-budget plans, so it does not rank models by profit alone.

User surveys: its best result

SurveyBench asks a model to plan a user survey and then read its results without falling for planted traps, such as a filtered base or a sample of current customers only. It is our newest benchmark, with eight models so far. Haiku 5.5 rated 1,412, third behind GPT-6.1 Sol (1,650) and Claude Opus 5.5 (1,646), and far ahead of Mistral Large 4 (815), Gemini 3.1 Pro (810) and Haiku 4.5 (465). Across all 12 tasks it gave every number correctly and handled 96% of the traps, for $0.23 in total. Its survey plans were sound for 7 of 12 tasks, but its analyses for only 3: most of the rest drew at least one finding the responses did not support.

Speed: fast to write, slow to start

Artificial Analysis measures Haiku 5.5 writing 235 tokens a second at max effort and 137 at medium, which is quick. Chess shows a different side. With three minutes a game it made a typical move in 6.2 seconds and reached 557 ladder Elo, above Haiku 4.5's 491. But at the 10-second Lightning clock it lost its first four games on time at every setting we tried, including with thinking off, and at the 60-second Bullet clock it only finished games with thinking off (310 Elo). Even then, a typical move took 1.7 seconds, against 1.3 for Haiku 4.5. These runs were on launch day, when providers are busiest, so we will re-run them.

In the playground: as good as GPT-6 Luna, at twice the price

We also gave Haiku 5.5 every playground test: short visual and interactive tasks such as games, simulations, drawings, landing pages and business pages, each a single self-contained file. So that it had something to be measured against, we gave the same 18 tests to GPT-6 Luna, which costs the same per token. They join ten other models already on each page. We played every answer and marked each test's checks by hand.

Haiku 5.5 is cheap here: all 18 answers cost $0.11, against $0.30 for DeepSeek V4.1 Flash and $6.09 for Claude Opus 5.5. But Luna did the same for $0.05, and was the cheapest answer on every test. On the checks, the two are level: each passed 85%. Haiku 5.5 passed every check on 10 tests and Luna on 8, and they split the tests they did not tie, five each. Haiku 5.5 was quicker, the fastest answer on 9 tests against Luna's 5.

18 playground tests: what each model passed, and what it cost

Share of our hand-marked checks passed across all 18 tests (a partial pass counts half), against the total cost of the answers, counting every try.

75%80%85%90%95%100% $0.01$0.1$1$10$100 Cost of all 18 answers (log scale) Checks passed GPT-6.1 Sol Claude Opus 5.5 Claude Fable 5.1 Gemini 3.8 Flash Muse Spark 1.3 Grok 4.7 Kimi K3 Qwen3.8 Max (0902) DeepSeek V4.1 Flash Mistral Large 4 Claude Haiku 5.5 GPT-6 Luna

Checks test whether an answer works, not how polished it is: our picks were more polished than Haiku 5.5's even where both passed every check.

Haiku 5.5 against GPT-6 Luna, test by test

Share of each test's hand-marked checks passed (a partial pass counts half). Haiku 5.5 scored higher on 5 tests, Luna on 5, and they tied on 8.

Claude Haiku 5.5GPT-6 Luna
Analogue clock
100% · 50%
Unicorn
83% · 33%
World map
62% · 38%
Game of Life
100% · 88%
Pricing table
100% · 88%
Animated logo
100% · 100%
Category page
100% · 100%
Chess
100% · 100%
Flappy Bird
100% · 100%
Landing page
83% · 83%
Product page
100% · 100%
Solar system
100% · 100%
Spinning hexagon
100% · 100%
Newsletter email
80% · 90%
The Starry Night
62% · 75%
Pelican on a bicycle
67% · 83%
Jelly effect
33% · 83%
Newton's cradle
33% · 100%

Haiku 5.5 was the more careful of the two on tests with rules to get exactly right: the analogue clock (Luna's hands turn about the wrong point), Conway's Game of Life (Luna's glider gun is drawn wrongly), the pricing table (Luna's switch never draws) and the two drawings of a unicorn and a world map. Luna did better where Haiku 5.5 is weakest: physics and facts. Haiku 5.5's Newton's cradle never transfers the swing and its jelly is a sliding ball, where Luna's work; its newsletter email invented product facts, including "made in small batches in the UK", where Luna's kept to the brief.

Two of the tests, with Haiku 5.5 next to the three models it is most often weighed against:

Draw a unicorn

Haiku 5.5 drew the whole animal; GPT-6 Luna drew only a head, with no legs or body.

Claude Haiku 5.5's answer: Draw a unicorn
Claude Haiku 5.5 · the new model
2.5 of 3 checks · $0.0017 · 16 s
GPT-6 Luna's answer: Draw a unicorn
GPT-6 Luna · the same price
1 of 3 checks · $0.0009 · 18 s
DeepSeek V4.1 Flash's answer: Draw a unicorn
DeepSeek V4.1 Flash · similar speed and intelligence
1.5 of 3 checks · $0.0175 · 67 s
Gemini 3.8 Flash's answer: Draw a unicorn
Gemini 3.8 Flash · similar intelligence
1.5 of 3 checks · $0.0356 · 74 s

Each model's single answer to the same prompt, drawn as SVG and rendered as it arrived. See all twelve answers, live, with our notes.

Pelican riding a bicycle

Here Luna did better: both feet on the pedals and a complete bicycle, where Haiku 5.5's bicycle has no pedals.

Claude Haiku 5.5's answer: Pelican riding a bicycle
Claude Haiku 5.5 · the new model
2 of 3 checks · $0.0013 · 13 s
GPT-6 Luna's answer: Pelican riding a bicycle
GPT-6 Luna · the same price
2.5 of 3 checks · $0.0012 · 20 s
DeepSeek V4.1 Flash's answer: Pelican riding a bicycle
DeepSeek V4.1 Flash · similar speed and intelligence
2 of 3 checks · $0.0061 · 52 s
Gemini 3.8 Flash's answer: Pelican riding a bicycle
Gemini 3.8 Flash · similar intelligence
2.5 of 3 checks · $0.0388 · 92 s

Each model's single answer to the same prompt, drawn as SVG and rendered as it arrived. See all twelve answers, live, with our notes.

Neither model is our pick on any test: the picks, mostly from GPT-6.1 Sol and Claude Opus 5.5, were more polished even where Haiku 5.5 passed every check. Every answer, with our notes, is on Claude Haiku 5.5's playground page.

Third-party benchmarks

  • Artificial Analysis (reported 7 October): Intelligence Index 43.4 at max effort, 41.2 at xhigh, 37.8 at high and 34.5 at the default medium setting, against 16.9 for Haiku 4.5 and 38.1 for GPT-6 Luna. Artificial Analysis calls it "notably fast, however very verbose": it used about 162,000 output tokens per Intelligence Index task at max effort, about three times GPT-6 Luna. On its AA-Omniscience test it answered fewer questions correctly than Luna or Gemini 3.8 Flash, but made things up less often (a 40% hallucination rate against 77% and 55%). Artificial Analysis says two of its figures are provisional: its AutomationBench-AA score, which a pre-release over-refusal problem probably lowered, and its cost figures, which do not yet include the higher price above 100,000 tokens.
  • Pending: Arena, tau2-bench, BFCL, OpenHands, STATE-Bench, Vending-Bench 2, the Vectara Hallucination Leaderboard, SimpleQA Verified and the professional-work sources (APEX-Agents, GDP.pdf, Remote Labor Index).

What people are saying about Claude Haiku 5.5

Reactions from launch day, on X, blogs and Hacker News. Quotes are as posted; each card links to the original.

How launch day read: 78% positive, 9% mixed, 13% negative of the 99 posts and comments that took a side

X66 of 202 top posts
Hacker News33 of 57 comments

NegativeMixedPositive

On 8 October we read the top posts on X for "Haiku 5.5" and "Claude Haiku 5.5", and every top-level comment on the Hacker News launch thread. We labelled each one positive, mixed, negative or neutral. News, jokes, questions, and posts from Anthropic, its staff and its launch partners count as neutral or are left out of the bars. Labels are by Claude Opus 5.5 against a written rubric, in a single pass; we have not yet checked a sample by hand. X ranks top posts by engagement, so its sample leans to the loudest voices, and on X most posts repeated Anthropic's launch claims. Every labelled post.

On X the reaction was overwhelmingly warm, mostly about the price cut and Anthropic's benchmark charts. The most-shared critical posts were hands-on tests, and they made the same point as our results: at the same price, GPT-6 Luna often did the job faster and for fewer tokens. A selection, with how we read each one:

Artificial Analysis@ArtificialAnlys

Anthropic has released Claude Haiku 5.5, scoring 43 on the Artificial Analysis Intelligence Index - up 26 points one year after the last Haiku release

7 Oct 2026 · View on XNeutral
Layton Gott@Layton_Gott

Anthropic somehow made Haiku MUCH smarter... then made it ~75% cheaper to run. […] Anthropic's cheap, fast model literally beats its old flagship on BOTH knowledge-work benchmarks.

7 Oct 2026 · View on XPositive
Jason Lemkin@jasonlk · SaaStr

Haiku 5.5 in the wild on SaaStr AI apps (out today):
- 1.5x faster
- 85% cheaper
- real reasoning (much more useable) […] Not all AI is inflationary

7 Oct 2026 · View on XPositive
Simon Willisonsimonwillison.net
Blog

If your workloads fit in 100,000 tokens, Haiku is the same price as Luna and reports higher benchmark scores. Above 100,000 tokens, Luna looks like a much better deal.

7 Oct 2026 · Read the postMixed
Chris Parmerchriddyp · Plotly
Hacker News

9x cheaper than Haiku 4.5 and 2 letter grades better. […] Compared to OpenAI: GPT-6 Luna did a bit better and was about 30% the cost of Haiku 5.5.

7 Oct 2026 · Read the commentMixed
Jacob Millerjjcm · diffui.ai
Hacker News

I was curious if this smaller model was good enough for complex UI. It was not. […] Very fast though, and likely best used for small subagent tasks / tightly scoped work.

7 Oct 2026 · Read the commentMixed
Haider.@haider1

WTF is going on with Haiku 5.5 at max reasoning effort it's more token-hungry than opus 5.5 max and uses over 2x more than fable 5.1 max

7 Oct 2026 · View on XNegative
BridgeMind@bridgemindai

Claude Haiku 5.5 just dropped and it completely fails the BridgeBench lava lamp test. […] GPT 6 Luna is over 6x faster and 3x cheaper.

7 Oct 2026 · View on XNegative
Hacker News727 points · our summary of 57 comments

Fans welcomed a Haiku that matches GPT-6 Luna's price and beats it on Anthropic's charts, and early testers praised its speed. Critics said the price step at 100,000 tokens comes too soon, that it burns tokens, and that their first tests were weak. Many comments were about the new API credits for subscribers instead.

7–8 Oct 2026 · Read the thread19 positive · 5 mixed · 9 negative

Claude Haiku 5.5 vs GPT-6 Luna, Gemini 3.8 Flash and DeepSeek V4.1 Flash

ModelWhy comparePrice*Tokens a secondAA IntelligenceSpring Prompt overall
Claude Haiku 5.5—$0.2023543.444
GPT-6 Luna compareThe same price$0.2012738.141
DeepSeek V4.1 Flash compareSimilar intelligence and speed$0.5221739.558
Gemini 3.8 Flash compareSimilar intelligence$1.5014740.961
Claude Haiku 4.5 compareThe previous Haiku$2.009716.98
Claude Sonnet 5.5 compareAnthropic's next step up$4.0013456.064

*List price per million tokens, blended at three input tokens to one output, on 8 October, for prompts up to 100,000 tokens. Speed and intelligence are from Artificial Analysis, at each model's best setting: for Haiku 5.5, max effort. The small figures show each model's difference from Claude Haiku 5.5: green where it is better, red where it is worse.

For most business text, GPT-6 Luna is the better buy at the same price: it invented fewer claims, planned better and used fewer tokens. DeepSeek V4.1 Flash and Gemini 3.8 Flash cost more but score well above both on our overall rating. If you are moving up from Haiku 4.5, Haiku 5.5 is better on everything we measure and costs a tenth as much, but test it against Luna on your own prompt before you settle.

Is Claude Haiku 5.5 good? Who should use it, and for what

  • Teams already on Haiku 4.5. It is better on every benchmark we run, and much cheaper per token.
  • Small, self-contained coding and front-end tasks. It passed every check on 10 of our 18 playground tests for under a penny each, especially where exact rules matter, though GPT-6 Luna matched it overall for half the price.
  • Extracting structured data from images and feeds. It rarely invented a product attribute and caught every conflict between feed and photo.
  • Not for customer-facing copy without a fact check. 60% of its product listings contained a claim the data did not support.
  • Not where every answer must arrive within a second or two. Even with thinking off, its typical reply took 1.7 seconds in our chess runs.
  • Watch long prompts. Above 100,000 tokens the price rises fivefold, which changes the sums for long documents.

As always, these benchmarks aren't your data, so run the prompt you actually use on Haiku 5.5 and the alternatives before switching, and compare the cost of each answer. Try it against GPT-6 Luna side by side in the playground, and if it would write your catalogue, our catalogue feed diagnostic measures how often it invents attributes on your real feed.

What we'll update

We'll publish a week-one verdict once Arena, tau2-bench and the hallucination leaderboards have results, and re-run BulletBench away from launch-day traffic. We may also run our benchmarks at high effort, which Anthropic recommends for knowledge work. Until then, the Claude Haiku 5.5 model page has every result as it lands.

Cheapest model that passes

On our benchmarks that define a pass for each task: what one passing answer cost in our run, failed attempts included, and whether a cheaper model passed.

  • CatalogBench · cost per 1,000 accepted listings
    Claude Haiku 5.5: $2.57 (79 of 168 attempts), 2 of 27. Cheapest: GPT-6 Luna at $0.70 (114 of 168 attempts). It passed at least as often.
  • DeckBench · cost per presentable deck
    Claude Haiku 5.5 passed 2 of 6 tasks, with no estimate: failed-call costs incomplete. Cheapest: GPT-6.2 Sol at $0.09 (5 of 6 tasks).
  • SurveyBench · cost per sound survey task
    Claude Haiku 5.5: $0.17 (1 of 9 tasks), 4 of 13. Cheapest: GPT-6 Luna at $0.03 (3 of 9 tasks). It passed at least as often.

Cost per attempt divided by the share that passed, at the run date's prices, excluding judging. Each model at its cheapest setting per passing answer. With a handful of tasks, a pass or two either way is within chance. No estimate where a run had missing outputs, because their cost may not have been recorded.

Quick answers

How much does Claude Haiku 5.5 cost?

Anthropic's list price is $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. A longer prompt is billed at $0.50 and $2.50 for the whole request. That is a tenth of Claude Haiku 4.5's price for short prompts and half for long ones.

What is Claude Haiku 5.5's context window?

1 million tokens, with up to 128,000 tokens of output (300,000 on the Message Batches API with a beta header). Haiku 4.5 had 200,000 tokens.

Does Claude Haiku 5.5 think before it answers?

Yes, by default. Adaptive thinking is on at medium effort, and there are five effort settings from low to max. Anthropic's headline benchmark scores are at max effort, not the default.

How good is Claude Haiku 5.5?

Artificial Analysis gives it 43.4 on its Intelligence Index at max effort and 34.5 at the default medium setting, against 16.9 for Haiku 4.5. On Spring Prompt's business benchmarks it is mid-table among models available today, and behind GPT-6 Luna, which costs the same.

Does Claude Haiku 5.5 make things up?

Less than Haiku 4.5, but still often. On CatalogBench, 60% of its product listings contained at least one claim the product data did not support, against 84% for Haiku 4.5 and 12.5% for GPT-6 Luna.

Claude Haiku 5.5 or GPT-6 Luna?

They cost the same per token. GPT-6 Luna beat Haiku 5.5 on our product listing, deck and ad-planning benchmarks and used fewer tokens per task. Haiku 5.5 writes faster and scores higher on the Intelligence Index at max effort.

Release dates: every Anthropic model release · new AI models released this week

Will Claude Haiku 5.5 work for your product?

Benchmarks aren't your task. Paste the prompt your product uses and run it beside the models you could use instead, with every answer checked and costed.

New here? Your email starts a free trial. No credit card needed.

More research
All posts →