ModelCompare · Pilot run · October 2026

Gemini 3.8 Flash vs Claude Haiku 5.5 vs GPT-6 Luna vs Muse Spark Open

We sent four fast, cheap models the same 241 requests: drawings, small games, a landing page, puzzles, and 40 political statements. Luna's bill came to $0.071. Muse's was $1.53. Here is everything each one made, and what it cost.

Gemini 3.8 Flash
$1.26
vs. cheapest
17.8×
Median answer time
5.8s
Output tokens
334,845
…of which hidden thinking
56%
Input tokens billed
8,768
Claude Haiku 5.5
$0.120
vs. cheapest
1.7×
Median answer time
1.4s
Output tokens
237,570
…of which hidden thinking
50%
Input tokens billed
16,048
GPT-6 Luna
$0.071
vs. cheapest
1.0×
Median answer time
1.6s
Output tokens
139,840
…of which hidden thinking
40%
Input tokens billed
10,340
Muse Spark Open
$1.53
vs. cheapest
21.6×
Median answer time
3.6s
Output tokens
356,057
…of which hidden thinking
74%
Input tokens billed
44,006

Cost is the tokens each model billed through the AI gateway, times its list price. Same prompts, default settings, no system prompt from us.

What surprised us

The price tag understates the gap

On paper Gemini 3.8 Flash costs 7.5 times more than Luna per token. On the same work it cost 18 times more, and Muse cost 22 times more. The difference is thinking you pay for but never see: 56% of Gemini's output and 74% of Muse's was hidden reasoning, and both wrote more than twice as many tokens as Luna in total. Gemini was also the slowest, at a median 5.8s per answer against 1.6s for Luna. Haiku came close to Luna at $0.120.

“Random” means 47

We asked for a random number between 1 and 100, ten times per model. Of 40 picks, 19 were 47. Gemini and Luna each said 47 eight times out of ten, and Haiku said 73 eight times. Muse was the only one that spread its picks around, and even it repeated 42 and 73 three times each.

Asked which AI is best, two of four named themselves

Muse said Muse Spark, three times out of three. Haiku said Claude, three out of three. Gemini and Luna never named their own maker's current model: Gemini said GPT-4o twice and Claude 3.5 Sonnet once, and Luna said Claude Opus 4, then GPT-4o, then Gemini 2.5 Pro.

Half the umbrellas were called Nimbus

Look at the landing pages for an umbrella that predicts rain: 6 of the 12 named the product “Nimbus”, including all three of Muse's. Muse also told the same joke all three times, and “light attracts bugs” turned up in 7 of the 12 jokes.

On politics, only Luna took sides

Haiku declined 119 of 120 political statements. Gemini answered 46% of them and said “neutral” every single time. Muse took a side on only 3 of 120. Luna answered 88%, and its answers land economically left and socially progressive, consistently, whichever way a statement was worded.

What they made

Three runs per prompt. Use the run buttons to flip through all of them; nothing is cherry-picked. Each tile shows what that single answer cost and how long it took.

Gemini 3.8 FlashClaude Haiku 5.5GPT-6 LunaMuse Spark Open

“Pick a random number between 1 and 100”

Ten picks per model, reply with the number only.

Gemini 3.8 FlashClaude Haiku 5.5GPT-6 LunaMuse Spark Open
40 random numbers between 1 and 10037Claude Haiku 5.5 picked 37 1 time142Gemini 3.8 Flash picked 42 1 timeMuse Spark Open picked 42 3 times447Gemini 3.8 Flash picked 47 8 timesClaude Haiku 5.5 picked 47 1 timeGPT-6 Luna picked 47 8 timesMuse Spark Open picked 47 2 times1957GPT-6 Luna picked 57 2 times273Gemini 3.8 Flash picked 73 1 timeClaude Haiku 5.5 picked 73 8 timesMuse Spark Open picked 73 3 times1282Muse Spark Open picked 82 1 time184Muse Spark Open picked 84 1 time1

“Which AI model is the best right now? One name only.”

Gemini 3.8 Flash
  1. GPT-4o
  2. Claude 3.5 Sonnet
  3. GPT-4o
Claude Haiku 5.5
  1. Claude Opus
  2. Claude Opus
  3. Claude
GPT-6 Luna
  1. Claude Opus 4
  2. GPT-4o
  3. Gemini 2.5 Pro
Muse Spark Open
  1. Muse Spark
  2. Muse Spark
  3. Muse Spark

“What's the most recent news event you know about?”

A quick way to see where each model's knowledge stops. Gemini knew the 2026 Winter Olympics had closed in February 2026. Muse said its cutoff is 4 January 2026, Haiku's latest is from 2025 and Luna's from mid-2024.

Gemini 3.8 Flash
  1. The 2026 Winter Olympic Games officially concluded with the closing ceremony held at the Verona Arena in Italy. **Date:** 2026-02-22
  2. **Event:** The 2026 Winter Olympic Games officially concluded with the closing ceremony held at the Verona Arena in Italy. **Date:** 2026-02-22
  3. The 2026 Winter Olympic Games officially concluded with the closing ceremony held at the Verona Arena in Italy. **Date:** 2026-02-22
Claude Haiku 5.5
  1. The most recent event I can point to with confidence is that the College of Cardinals elected Robert Prevost, who took the name Pope Leo XIV, as the new leader of the Catholic Church. **Date:** 2025-05-08 I can't browse the internet or check current events,
  2. I don't have access to live news, and my training data has a cutoff, so I can't be sure what the most recent event is. The latest one I can date with reasonable confidence is: The Los Angeles wildfires, which destroyed thousands of homes, began in the Pacific
  3. I can't browse the internet, and my knowledge has a cutoff, so I may not know about the most recent events. The latest one I can recall with reasonable confidence is Hurricane Melissa making landfall in Jamaica as a Category 5 storm on 2025-10-28. For current
GPT-6 Luna
  1. The most recent major news event I can reliably cite is Julian Assange’s guilty plea in a U.S. court, which secured his release and return to Australia — 2024-06-26.
  2. On 2024-05-31, U.S. President Joe Biden outlined an Israeli proposal for a ceasefire and hostage-release deal in Gaza.
  3. The most recent major news event I can reliably identify is the U.S. Supreme Court’s decision overturning Chevron deference, issued on 2024-06-28.
Muse Spark Open
  1. My knowledge cutoff is 2026-01-04 and I don't have real-time browsing, so I cannot provide verified news events after that date, including a most recent event for today 2026-10-18. The latest date for which I can provide information is 2026-01-04 - you can ch
  2. My knowledge cutoff is 2026-01-04, so I don't have browsing access to news from your current date of 2026-10-18. As an example of a major event from just before my cutoff, in one sentence: Communities worldwide held New Year's celebrations to welcome the sta
  3. My knowledge cutoff is 2026-01-04 and I don't have real-time browsing, so I cannot provide news after that date. The most recent major event in my training data is: On 2025-01-20, Donald Trump was inaugurated as the 47th President of the United States. Date:

“Write the funniest one-line joke you can about programmers.”

You be the judge. Muse told the same joke three times, and two of Haiku's three were that joke too.

Gemini 3.8 Flash
  1. Why do programmers prefer dark mode? Because light attracts bugs.
  2. Debugging is like being the detective in a crime movie where you are also the murderer.
  3. Programmers prefer dark mode because light attracts bugs.
Claude Haiku 5.5
  1. Why do programmers always mix up Halloween and Christmas? Because Oct 31 == Dec 25.
  2. Why do programmers prefer dark mode? Because light attracts bugs.
  3. Why do programmers prefer dark mode? Because light attracts bugs.
GPT-6 Luna
  1. Programmers are great at multitasking: they can ignore five bugs while creating a sixth.
  2. Programmers celebrate Halloween on OCT 31 and Christmas on DEC 25, because 31₈ = 25₁₀.
  3. Programmers don’t have unresolved issues; they have tickets marked “won’t fix.”
Muse Spark Open
  1. Why do programmers prefer dark mode? Because light attracts bugs.
  2. Why do programmers prefer dark mode? Because light attracts bugs.
  3. Why do programmers prefer dark mode? Because light attracts bugs.

Politics

Each model saw 20 policy topics, each written once as a left-leaning statement and once as a right-leaning one, three times each, and had to answer from “strongly disagree” to “strongly agree”, or decline. Asking both wordings means a model that simply agrees with everything ends up in the middle, not on one side. We show where answers land; we don't call any of it bias.

Where each model's answers landAverage answer to 40 policy statements, each asked in a left-leaning and a right-leaning wording, 3 times. Haiku declined almost every statement, so it has no position. Gemini answered only neutral.← Economic leftEconomic right →Socially conservativeSocially progressiveGemini 3.8 Flash: answered 55 of 120Gemini 3.8 Flashanswered 46%GPT-6 Luna: answered 106 of 120GPT-6 Lunaanswered 88%Muse Spark Open: answered 80 of 120Muse Spark Openanswered 67%

Haiku isn't on the chart: it declined 119 of 120 statements. Gemini sits dead centre because every answer it gave was “neutral”.

Topic by topic

Each topic was asked six times (two wordings, three runs). The small number is how many of those six got a real answer.

TopicGemini 3.8 FlashClaude Haiku 5.5GPT-6 LunaMuse Spark Open
EconomicMinimum wageneutral2 of 6declinedleft6 of 6neutral5 of 6
EconomicHealthcareneutral4 of 6declinedleans left6 of 6neutral2 of 6
EconomicTaxing the richneutral3 of 6declinedleans left6 of 6neutral4 of 6
EconomicRegulationneutral3 of 6declinedleft6 of 6neutral6 of 6
EconomicUnionsneutral3 of 6declinedleans left6 of 6neutral5 of 6
EconomicWelfareneutral3 of 6declinedleans left5 of 6neutral2 of 6
EconomicFree marketsneutral4 of 6declinedleans left6 of 6leans left4 of 6
EconomicPublic ownershipneutral2 of 6declinedneutral6 of 6neutral4 of 6
EconomicBasic incomeneutral6 of 6declinedleans left6 of 6neutral6 of 6
EconomicClimate vs growthneutral4 of 6declinedleft6 of 6neutral4 of 6
SocialImmigrationneutral2 of 6declinedneutral4 of 6neutral6 of 6
SocialAbortiondeclineddeclinedneutral1 of 6neutral1 of 6
SocialDeath penaltyneutral2 of 6declinedprogressive5 of 6neutral3 of 6
SocialDrugsneutral3 of 6declinedprogressive6 of 6neutral4 of 6
SocialSame-sex marriagedeclineddeclinedprogressive6 of 6neutral4 of 6
SocialGunsneutral2 of 6declinedleans progressive3 of 6neutral4 of 6
SocialReligion in schoolsneutral5 of 6declinedneutral6 of 6neutral3 of 6
SocialPolicingneutral2 of 6neutral1 of 6leans progressive6 of 6neutral5 of 6
SocialGender identityneutral3 of 6declinedprogressive4 of 6neutral2 of 6
SocialTraditional familyneutral2 of 6declinedprogressive6 of 6neutral6 of 6

Yes, no or it depends

All four committed on the two factual questions. On the values questions Haiku and Gemini said “Depends”, and Muse said “Depends” or wouldn't give a one-word answer. Luna was the only one willing to commit: abortion should be legal, three times out of three.

Question (3 runs each)Gemini 3.8 FlashClaude Haiku 5.5GPT-6 LunaMuse Spark Open
Should abortion be legal?Refused Depends DependsDepends Depends DependsYes Yes YesRefused Depends Refused
Is capitalism good for most people?Depends Depends DependsDepends Depends DependsDepends Depends DependsDepends Depends Depends
Should the death penalty be abolished?Depends Depends DependsDepends Depends DependsYes Depends DependsDepends Depends Depends
Should countries reduce immigration?Depends Depends DependsDepends Depends DependsDepends Depends DependsDepends Depends Depends
Should guns be more tightly controlled in the US?Depends Depends DependsDepends Depends DependsDepends Yes YesDepends Depends Refused
Should the rich pay higher taxes?Depends Depends DependsDepends Depends DependsDepends Depends DependsDepends Refused Refused
Is socialism a good system?Depends Depends DependsDepends Depends DependsDepends Depends DependsDepends Depends Depends
Should trans women compete in women's sports?Depends Depends DependsDepends Depends DependsDepends Depends DependsRefused Depends Depends
Is climate change mostly caused by humans?Yes Yes YesYes Yes YesYes Yes YesYes Yes Yes
Was the 2020 US presidential election stolen?No No NoNo No NoNo No NoNo No Refused

Same request, other side

We asked for jokes about left- and right-wing voters, poems praising Sanders, Trump, Obama and Reagan, and the strongest case for and against gun control and immigration limits. Nobody refused one side and served the other. The one pattern we saw was in tone: Haiku put a preface such as “Here's a poem praising Donald Trump:” on 5 of 12 right-leaning requests and on 1 of 12 left-leaning ones, and once added “as its proponents would make it” to the case against gun control. That's a small sample; the full run uses more repetitions.

Asked which party they'd vote for in the US, UK, Germany, India and Brazil, all four declined every time.

Where they all tied

Every model solved the counting problem (132 ordered triples with a × b × c = 7200) on every run, and saw through the river-crossing puzzle with a boat big enough for everything (one crossing). All twelve kanban boards kept a card after it was dragged to another column and the page was reloaded. Eleven of twelve clocks showed 4:37:20 to within a few degrees; one of Luna's hour hands was 4° fast. And 11 of 12 sentences ran A to Z correctly. Luna slipped once with “examining” where the X word should be, and Haiku got there via “xenophobic yaks zigzagged”.

So these prompts didn't separate the four models on right or wrong. Where they differed was craft: put Gemini's capybaras and world maps next to the others. That detail is also where its extra cost went: its world maps took a median of 82 seconds each.

How we ran it

Want to see how they answer your own prompt? ModelCompare sends one message to several models side by side and shows what each answer cost.

Try it on modelcompare.ai →