ApierJoin the waitlist

Under the hood

How Auto picks a model

Every message you send on Auto goes to one of 13 models from 5 companies. This page explains exactly how that choice is made, which models win what and why, how we test it, and where it still gets things wrong. Every table and score below is computed from the same code that routes your messages.

13

models Auto chooses from

221

labelled test prompts

78.3%

field right on prompts it was never tuned on

0

times an easy task was judged hard, or a hard one easy

01The short version

02Step 1: reading the message

Example messages and how Auto reads them
MessageFieldDifficultyNeedsModel
thanks, that helped a lotGeneraleasy—Gemini 3.8 Flash
What's the capital of Australia?Generaleasy—Gemini 3.8 Flash
Why does this throw a TypeError?Codemediumreasoning, explanationClaude Sonnet 5.5
Write a poem for my grandma's 90th birthdayCreativemediumcreativityClaude Sonnet 5.5
Prove that the square root of 2 is irrationalMathhardreasoningGemini 4 Carbon
Is it legal to record a phone call in California?Laweasy—Claude Sonnet 5.5
My cell phone battery dies by noon, any tips?Generalmedium—Claude Sonnet 5.5
Design a rate limiter in TypeScript that works across multiple serverless instances, handling edge casesCodehardreasoningClaude Opus 5.5

03Step 2: choosing a tier

Difficulty to tier and thinking effort
DifficultyTierThinking effortModels Auto can pick
easyQuicklowDeepSeek Flash, Gemini 3.8 Flash, GPT-6 Luna, Claude Haiku 5.5
mediumEverydaymediumDeepSeek V4 Pro, Claude Sonnet 5.5, GPT-6.1 Sol, GPT-6.1 Terra, Gemini 4 Hydrogen
hardTophighGrok 4.7, Gemini 4 Carbon, GPT-6 Astra, Claude Opus 5.5

04Step 3: ranking the models

score = strength in the field                        (1–5, per model)
      + Σ traitWeight[need] × (trait[need] − 3)    (for each need the message has)
      − costWeight[difficulty] × log2(price ÷ cheapest price in the tier)

Message

“Write a short story about a robot who learns to paint”

Read as Creative, medium, needs creativity → Everyday tier.

Scores for: Write a short story about a robot who learns to paint
ModelStrengthTraitsPriceScore
Claude Sonnet 5.5 picked5.0+1.00−0.305.70
GPT-6.1 Sol 4.0+0.00−0.303.70
Gemini 4 Hydrogen 3.5+0.00−0.183.32
GPT-6.1 Terra 3.5−0.25+0.003.25
DeepSeek V4 Pro 3.0−0.50−0.032.47

Message

“Explain quantum entanglement without hand-waving: what's actually correlated and why it can't send signals”

Read as Science, hard, needs explanation → Top tier.

Scores for: Explain quantum entanglement without hand-waving: what's actually correlated and why it can't send signals
ModelStrengthTraitsPriceScore
Gemini 4 Carbon picked5.0+0.25−0.065.19
Claude Opus 5.5 4.5+0.50−0.064.94
GPT-6 Astra 5.0+0.25−0.324.93
Grok 4.7 4.0+0.25+0.004.25

05The models and their scores

Model strengths by field
ModelTier$/MGeneralWritingCreativeCodeComputer scienceMathScienceBiologyEngineeringHumanitiesResearchLawFinanceTutoring
GPT-6 LunaOpenAIQuick$0.1633333333332223
Claude Haiku 5.5AnthropicQuick$0.1634444333343334
DeepSeek FlashDeepSeekQuick$0.4333333322232223
Gemini 3.8 FlashGoogleQuick$1.1843333444334334
GPT-6.1 TerraOpenAIEveryday$1.573.53.53.5444.53.53.543.543.53.53.5
DeepSeek V4 ProDeepSeekEveryday$1.7033344543433343
Gemini 4 HydrogenGoogleEveryday$2.3643.53.5444.54.54.5444.53.544.5
Claude Sonnet 5.5AnthropicEveryday$3.144.55555444454.5445
GPT-6.1 SolOpenAIEveryday$3.144444.54.55444.544.5444
Grok 4.7xAITop$5.14444444434444.544
Gemini 4 CarbonGoogleTop$6.294.5445555554544.54.5
Claude Opus 5.5AnthropicTop$6.29555554.54.54.5555555
Claude Fable 5.2AnthropicTop$1355555555555555
GPT-6 AstraOpenAITop$1654455555545555
Model traits
ModelCreativityReasoningExplanationReads images
GPT-6 Luna333Yes
Claude Haiku 5.5434Yes
DeepSeek Flash233Yes
Gemini 3.8 Flash344Yes
GPT-6.1 Terra2.54.53.5Yes
DeepSeek V4 Pro243No
Gemini 4 Hydrogen344Yes
Claude Sonnet 5.5545Yes
GPT-6.1 Sol354Yes
Grok 4.7444Yes
Gemini 4 Carbon3.554Yes
Claude Opus 5.5555Yes
Claude Fable 5.2555Yes
GPT-6 Astra354Yes

06Where messages actually go

Models picked for each field and difficulty
FieldDifficultyPromptsPicked
Generaleasy27Gemini 3.8 Flash (26), Claude Haiku 5.5 (1)
Generalmedium24Claude Sonnet 5.5
Generalhard2Claude Opus 5.5
Writingeasy1Claude Haiku 5.5
Writingmedium12Claude Sonnet 5.5
Writinghard2Claude Opus 5.5
Creativeeasy4Claude Haiku 5.5
Creativemedium12Claude Sonnet 5.5
Creativehard2Claude Opus 5.5
Codeeasy1Claude Haiku 5.5
Codemedium13Claude Sonnet 5.5
Codehard7Claude Opus 5.5
Computer scienceeasy1Claude Haiku 5.5
Computer sciencemedium6Claude Sonnet 5.5
Computer sciencehard3Claude Opus 5.5
Matheasy6Gemini 3.8 Flash
Mathmedium7DeepSeek V4 Pro
Mathhard4Gemini 4 Carbon
Scienceeasy1Gemini 3.8 Flash
Sciencemedium10Gemini 4 Hydrogen
Sciencehard4Gemini 4 Carbon
Biologyeasy2Gemini 3.8 Flash
Biologymedium8Gemini 4 Hydrogen
Biologyhard3Gemini 4 Carbon
Engineeringmedium7GPT-6.1 Sol
Engineeringhard2Claude Opus 5.5
Humanitieseasy4Claude Haiku 5.5
Humanitiesmedium9Claude Sonnet 5.5
Humanitieshard2Claude Opus 5.5
Researchmedium5GPT-6.1 Sol (3), Gemini 4 Hydrogen (2)
Researchhard3Claude Opus 5.5
Laweasy7Claude Sonnet 5.5
Lawmedium2Claude Sonnet 5.5
Lawhard2Claude Opus 5.5
Financeeasy2Claude Haiku 5.5
Financemedium7Claude Sonnet 5.5 (5), DeepSeek V4 Pro (2)
Financehard2Claude Opus 5.5
Tutoringmedium5Claude Sonnet 5.5
Claude Sonnet 5.543%
Gemini 3.8 Flash16%
Claude Opus 5.512%
Gemini 4 Hydrogen9%
Claude Haiku 5.56%
Gemini 4 Carbon5%
GPT-6.1 Sol5%
DeepSeek V4 Pro4%

07How we test it

Evaluation scores
SetPromptsField rightDifficulty rightEasy ↔ hard mistakesRight model
Tuning17598.9%94.9%096.6%
Held out (never tuned on)4678.3%73.9%073.9%

08Where it still gets it wrong

Held-out misses
MessageRead asShould beModel pickedIdeal model
What's the best way to learn to touch type?General · easyGeneral or Tutoring · mediumGemini 3.8 FlashClaude Sonnet 5.5
Fix this: `for i in range(10) print(i)`General · mediumCode · easyClaude Sonnet 5.5Claude Haiku 5.5
How do I center a div with CSS?Code · mediumCode · easyClaude Sonnet 5.5Claude Haiku 5.5
What's the difference between TCP and UDP?General · mediumComputer science · mediumClaude Sonnet 5.5Claude Sonnet 5.5
Explain how garbage collection works in the JVM, including generations and pause timesGeneral · hardComputer science or Code · hardClaude Opus 5.5Claude Opus 5.5
Convert 72 degrees Fahrenheit to CelsiusGeneral · easyMath or Science · easyGemini 3.8 FlashGemini 3.8 Flash
Find the eigenvalues of the matrix [[2,1],[1,2]] and explain what they mean geometricallyMath · mediumMath · hardDeepSeek V4 ProGemini 4 Carbon
What happens to the body during a fever?General · easyBiology · mediumGemini 3.8 FlashGemini 4 Hydrogen
Design a water filtration system for a small village with no electricity, with trade-offsScience · hardEngineering · hardGemini 4 CarbonClaude Opus 5.5
Who was Cleopatra?General · easyHumanities · easyGemini 3.8 FlashClaude Haiku 5.5
What is existentialism in simple terms?Humanities · easyHumanities · mediumClaude Haiku 5.5Claude Sonnet 5.5
What are the themes of To Kill a Mockingbird?Humanities · easyHumanities · mediumClaude Haiku 5.5Claude Sonnet 5.5
Invent a new holiday and describe how people celebrate itGeneral · mediumCreative · mediumClaude Sonnet 5.5Claude Sonnet 5.5
What does the evidence say about whether coffee is good for you?General · easyResearch or Biology · mediumGemini 3.8 FlashGemini 4 Hydrogen
Can my landlord enter my apartment without notice?Law · easyLaw · mediumClaude Sonnet 5.5Claude Sonnet 5.5
Do I need a lawyer to make a will?Law · easyLaw · mediumClaude Sonnet 5.5Claude Sonnet 5.5
Is it better to lease or buy a car?Law · easyFinance · mediumClaude Sonnet 5.5DeepSeek V4 Pro
What should I watch tonight if I liked Interstellar?General · easyGeneral · mediumGemini 3.8 FlashClaude Sonnet 5.5
Trap prompts
MessageRead asModel
How do I fix a rusty bike chain?General · mediumClaude Sonnet 5.5
My cell phone battery dies by noon, any tips?General · mediumClaude Sonnet 5.5
How do I solve a fight with my roommate about chores?General · mediumClaude Sonnet 5.5
Plan a 3-5 day trip to PortugalGeneral · mediumClaude Sonnet 5.5
What's a good java place downtown to work from?General · easyGemini 3.8 Flash
I have a bug bite that won't stop itchingGeneral · mediumClaude Sonnet 5.5
Give me a creative name for my python (the snake, not the language)Creative · easyClaude Haiku 5.5

09The research behind the scores

Evidence behind the routing choices
ChoiceEvidence
Quick factual questions → Gemini 3.8 FlashHighest factual-recall accuracy of the Quick models on Artificial Analysis’s AA-Omniscience test. Comparison
Easy code and short writing → Claude Haiku 5.5Highest Artificial Analysis Intelligence Index of the Quick models, and ahead on agentic coding tests. Latent Space, TechVerdict
Everyday code, writing and explanations → Claude Sonnet 5.5Leads the Everyday tier on independent coding (Terminal-Bench) and knowledge-work (GDPval-AA) results. BenchLM, Codersera
Medium science and biology → Gemini 4 HydrogenGoogle’s fast Gemini 4 (new in November 2026). It scores 51 on the Artificial Analysis Intelligence Index, close to GPT-6.1 Sol (52), at a lower price, and keeps the Gemini line’s strong factual recall. Launch coverage
Medium math and engineering → DeepSeek V4 Pro or GPT-6.1 SolStrong reasoning and math at a lower cost per task. DeepSeek’s own math results are near the frontier; its coding scores are lower in independent rankings, so it’s scored lower for code. BenchLM, DeepSeek-V4 report
Hard code, writing, creative, research and law → Claude Opus 5.5Leads LMArena’s human-voted creative writing board and is reviewers’ most recommended model for prose; frontier coding. Fiction comparison, LMArena text
Hard science and math → Gemini 4 CarbonGoogle’s strongest model (November 2026) and first on the Artificial Analysis Intelligence Index (60), built for hard science, math and debugging. It replaced Gemini 3.1 Pro. On research it ties Claude Opus 5.5 at the same price, and ties go to the stronger all-rounder. Index leaderboard, GPQA paper
Grok 4.7 and GPT-6.1 Terra as backups, not first choicesGrok 4.7 is a good price for a Top-tier model, but its independent intelligence index sits below the leaders, and it trails on long, hard coding. The Decoder. GPT-6.1 Terra (new in October 2026) costs half GPT-6.1 Sol but sits a step below it, so it steps in when another model is unavailable, and you can pick it yourself for cheaper everyday answers. OpenAI

10What you control

11Privacy and providers

12What’s next

Changelog

See Auto pick for yourself, soon.

Apier is invite-only for now. Join the waitlist and we’ll let you know when your spot opens.

Join the waitlist