Plexe
Plexe Models

Specialist models
for travel.

Small models that each do one job on travel data: group a hotel's rooms across suppliers, or find the place a traveller means from half a word. Each is built for its task, measured before every release, and fast enough to call live.

Room mapping benchmark · mlr-1

More accurate than the leading provider, in under a second.

The task: take one hotel's room names from every supplier and group the ones that are the same room. Benchmarked on 6,337 blind hotels from around the world and from live customer traffic, never used to train or tune the model, against one written definition of a room.

91.7F1 on same-room grouping, against 82.3 for the leading provider
95.1F1 on room types, against 61.3
0.66sper hotel, median, through the live API
$0.20per 1,000 requests
Table 1. mlr-1 against the leading room-mapping provider on 6,337 blind hotels, each with room names from at least two suppliers. Same room: offers that are one bookable room. Room type: rooms that differ only in beds or occupancy.
LevelSystemF1PurityCompleteness
Same roomLeading provider82.383.780.9
Same roommlr-191.794.589.1
Room typeLeading provider61.393.745.6
Room typemlr-195.196.094.1

B³ F1 over every supplier name. Purity: each card holds one room. Completeness: each room sits on one card. Labels follow our published room definition, made by two independent LLM referees with an adjudicator. Both systems received the same supplier room names. 81.6% of hotels resolve in under a second, timed end to end by the caller. 16 Nov 2026.

Against general models · 200 hotels

GPT-5.5 and GPT-6.1 match mlr-1's F1 at 17+ seconds a hotel: mlr-1 is 67× faster and 44× cheaper than GPT-6.1 at the same F1.

under 1s 60708090100 50 0.1s1s10s100s every name its own card · 73.6 GPT-5.5 · F1 94.1 · 44s per hotel · $132.11 per 1,000 hotels GPT-5.5 · 44s GPT-6.1 · F1 93.6 · 17s per hotel · $8.72 per 1,000 hotels GPT-6.1 · 17s gpt-oss-120b · F1 89.0 · 26s per hotel · $4.00 per 1,000 hotels gpt-oss-120b · 26s JEV (Typesafe) · F1 86.2 · 0.38s per hotel · $0.62 per 1,000 hotels JEV (Typesafe) · 0.38s DeepSeek V3.2 · F1 68.2 · 5s per hotel · $2.40 per 1,000 hotels DeepSeek V3.2 · 5s Qwen3 32B · F1 59.3 · 3s per hotel · $0.67 per 1,000 hotels Qwen3 32B · 3s Llama 3.3 70B · F1 52.1 · 2s per hotel · $1.39 per 1,000 hotels Llama 3.3 70B · 2s mlr-1 · F1 93.0 · 0.26s per hotel · $0.20 per 1,000 hotels mlr-1 93.0 F1 · 0.26s
Figure 1. General models on 200 hotels: B³ F1 against median seconds per hotel, log scale. Shaded: fast enough for live search. Swipe for the rest →
Table 2. Results on 200 hotels from the 1,000-hotel blind benchmark. Purity: each card holds one room. Completeness: each room sits on one card.
ModelF1PurityCompletenessSeconds per hotel$ per 1,000 hotels
GPT-5.594.192.196.244.0$132.11
GPT-6.193.691.396.017.3$8.72
mlr-193.094.291.90.26$0.20
gpt-oss-120b open89.096.682.425.8$4.00
JEV (Typesafe)86.292.180.90.38$0.62
DeepSeek V3.2 open68.258.781.54.7$2.40
Qwen3 32B open59.351.470.13.1$0.67
Llama 3.3 70B open52.139.078.51.6$1.39
mlr-15,000
JEV (Typesafe)1,613
Qwen3 32B1,493
Llama 3.3 70B719
DeepSeek V3.2417
gpt-oss-120b250
GPT-6.1115
GPT-5.58
Figure 2. Hotels resolved per $1. A million hotels: $200 with mlr-1, $132,110 with GPT-5.5.
Before / after

Same rates. Fewer cards.

American Inn by OYO Colorado City I-20, Colorado City, from the benchmark, resolved in 203 ms. Hover a room to see its names.

Today19 cards
  • Queen Bed Non-Smoking
  • 2 Double Beds Smoking
  • Queen Bed Smoking
  • 2 Double Beds Non-Smoking
  • Queen Bed Accessible
  • ROOM, QUEEN BED
  • ROOM, QUEEN BED, SMOKING
  • ROOM, 2 DOUBLE BEDS, SMOKING
  • ROOM, 2 DOUBLE BEDS
  • DOUBLE TWO DOUBLE BEDS
  • DOUBLE QUEEN BED NON SMOKING
  • DOUBLE QUEEN BED SMOKING
  • DOUBLE TWO DOUBLE BEDS NON SMOKING
  • DOUBLE Accessible-Queen Size Bed
  • Run of House
  • Standard room
  • Standard Quadruple room
  • Standard Double room
  • Standard Sextuple room
With Plexe8 rooms
  • Queen Room6 names
  • Room, 2 Double Beds6 names
  • Queen Room, Accessible2 names
  • Run of House1 name
  • Standard Room1 name
  • Standard Quadruple Room1 name
  • Standard Double Room1 name
  • Standard Sextuple Room1 name
Why it's hard

Room names are language, not data.

  • Signature Suite, Bathtub
  • Signature Suite, Sea View
  • Signature Suite Taj Club, Lounge Access
1 room
Three suppliers, one suite at Taj Lands End. Rules match patterns; wording like this breaks them.

1 in 5

names grouped differently by two labelling passes that followed the same written rules. Naming is a judgement.
Rule enginesDecision model
BuiltPer supplier, by handOnce, for every supplier
Taught byPatterns and code tablesA written definition of a room
New wordingBreaksReads it
New customerRebuilds the workClean rooms from the first call
Pricing

Pay per request.

One request maps one hotel's rooms from all its suppliers. Not priced per room or supplier.

Usage

$0.20 per 1,000 requests

  • No monthly fee or minimum
  • One rate at any volume, billed monthly
  • Live and batch API, room hierarchy, support
Request access

Enterprise

Custom

  • Volume pricing above 10 million requests a month
  • Dedicated GPU and an availability target
  • A shared support channel
Talk to us

A request is one successful API call; in a bulk upload, each hotel is one request. Errors and playground calls are free. Prices exclude taxes.

Travel

Starting with travel

Travel inventory reaches you from dozens of suppliers, each describing the same hotel, room and photo in its own way. Our first models resolve that, behind one API.

Use the models

Read the docs, try mlr-1 in your browser, or sign in with your API key.