Map of Taste

See how meaning maps across the menu.

Color is origin

100+languages

0.869search NDCG@10

4.7xdiet search vs. rivals

~200msp50 latency

Benchmarks

How Latimal compares

Built for food and benchmarked against the strongest general-purpose models.

#1of 10 on the FoodEval leaderboard0.819 avg across three categories

Overall

Category averages across the three tables below.

All tasks from FoodEval, the public food-embedding benchmark.
Search is reranked retrieval: each competitor's embeddings are reranked with bge-reranker-v2-m3, and Latimal runs its production /search endpoint.
Matching and classification: raw embeddings + cosine, scored the same way for every model.
Average: unweighted mean of the three category scores.

Overall: Latimal compared with seven embedding systems. Higher is better.
Task	Latimal	OpenAItext-embedding-3-large	Voyage AIVoyage 4 Large	CohereEmbed v4	AlibabaGTE-large	Nomic AINomic v1.5	BAAIBGE-M3	MicrosoftE5-large
Average3 categories	0.819	0.682	0.659	0.643	0.614	0.624	0.609	0.505
SearchProduction NDCG@10, 4 tasks	0.869	0.455	0.447	0.451	0.427	0.422	0.408	0.411
MatchingMean F1, 7 tasks	0.851	0.758	0.741	0.741	0.699	0.739	0.718	0.704
ClassificationMacro F1, 1 task	0.738	0.833	0.789	0.737	0.716	0.710	0.701	0.399

Matching

Best F1, 7 tasks.

Same-dish detection across cuisines, scripts, and noise levels.
Raw embeddings + cosine, no reranking.

Matching: Latimal compared with seven embedding systems. Higher is better.
Task	Latimal	OpenAItext-embedding-3-large	Voyage AIVoyage 4 Large	CohereEmbed v4	AlibabaGTE-large	Nomic AINomic v1.5	BAAIBGE-M3	MicrosoftE5-large
Average7 tasks	0.851	0.758	0.741	0.741	0.699	0.739	0.718	0.704
Indian cuisine	0.817	0.745	0.718	0.732	0.705	0.731	0.711	0.680
Global cuisine	0.867	0.828	0.783	0.829	0.695	0.732	0.716	0.716
Beverages	0.746	0.715	0.719	0.710	0.710	0.715	0.706	0.706
Bakery & desserts	0.755	0.735	0.715	0.691	0.682	0.684	0.684	0.688
Portion size	0.972	0.849	0.791	0.835	0.725	0.855	0.821	0.757
Noisy menu	0.916	0.685	0.640	0.667	0.672	0.750	0.674	0.648
Cross-lingual	0.886	0.748	0.820	0.721	0.707	0.707	0.717	0.731

Search

Production search, NDCG@10.

Latimal: production API, measured at the public API boundary. Reproducible with an API key.
Competitors: embedding + bge-reranker-v2-m3.

Search: Latimal compared with seven embedding systems. Higher is better.
Task	Latimal	OpenAItext-embedding-3-large	Voyage AIVoyage 4 Large	CohereEmbed v4	AlibabaGTE-large	Nomic AINomic v1.5	BAAIBGE-M3	MicrosoftE5-large
Average4 tasks	0.869	0.455	0.447	0.451	0.427	0.422	0.408	0.411
Food searchNDCG@10	0.938	0.590	0.590	0.589	0.572	0.564	0.552	0.554
Concept searchNDCG@10	0.809	0.405	0.392	0.391	0.374	0.357	0.336	0.328
Diet & allergen searchNDCG@10	0.802	0.172	0.161	0.165	0.135	0.132	0.132	0.136
Noisy searchNDCG@10	0.925	0.653	0.644	0.660	0.628	0.635	0.614	0.628

Diet & allergen search: 4.7x the best competitor.

Classification

Macro F1, 1 task. Linear probe on frozen embeddings, 26 menu classes.

Classification: Latimal compared with seven embedding systems. Higher is better.
Task	Latimal	OpenAItext-embedding-3-large	Voyage AIVoyage 4 Large	CohereEmbed v4	AlibabaGTE-large	Nomic AINomic v1.5	BAAIBGE-M3	MicrosoftE5-large
Cuisine classificationMacro F1	0.738	0.833	0.789	0.737	0.716	0.710	0.701	0.399

Measured on equal footing, June 2026. Full benchmarks on Hugging Face → FoodEval leaderboard →

One API, the whole menu

What a delivery platform needs to understand food.

Diners get search and recommendations that work across the full menu. Underneath, Latimal matches the same dish across restaurants.

POST /search

Semantic search

A diner types “something brothy and warming” and the right bowls rise to the top, whatever they’re called on the menu.

RamenPhoLaksaDan Dan Mian

POST /classify

Cuisine classification

Sort an item into its cuisine, from Levantine to Andean, even when the menu never spells it out.

LevantineAndeanSichuan

POST /match

Matching & price

The same dish hides under different names and prices across restaurants. Match them and line the prices up side by side.

Pad ThaiPhat Thaiผัดไทย

POST /suggest

Cart upsell

A cart of ramen and gyoza suggests what pairs with it. A craving for “cold & refreshing” answers across cuisines.

EdamameGreen TeaMochi

Integration

Up and running in three calls.

One REST API. Nothing to host.

Get your keySign up and grab a key from the dashboard.
Send a callOne POST with your items and a query.
Read resultsScored matches come back as JSON.

POST /search

{ "query": "dumplings",
  "corpus": ["Gyoza", "Pierogi", "Gazpacho", "Empanada"] }

// => Gyoza (0.95), Pierogi (0.93), Empanada (0.91)

Get started

Try it free for 14 days.

Pick a plan and start a 14-day trial with full API access. Your card is charged only after the trial ends.

Run the full API on a real menu before you pay.
Move from pilot to production on the same key, no rebuild.

Not sure which plan fits? Write to us and we'll help you pick the right one.

14days free

Full access to every endpoint on your plan for 14 days.
Cancel anytime before the trial ends, no charge.
Three plans from $59/mo. Pick the one that fits your volume.

See plans

Private by default

Your menus stay yours and are never reused.

Built for scale

p50 ~200ms per query, plus bulk endpoints.

Production-ready

Warm-failover standby, 99.5% uptime.

No infra to host

No GPUs and no infra to keep running.

Put your menu data to work.

Give your platform search and recommendations that read food the way diners do. Start with a 14-day free trial.

Get API key Read the docs

14: days free
#1: on FoodEval