CohereLanguage modelOpen weights

North-Small-Translate-1.0

North-Small-Translate-1.0 has no overall index value yet because there are too few independent benchmark results.

PopularityRank 670 of 875

0out of 100same as last month

Summary

North-Small-Translate-1.0 focuses Cohere Labs’ MoE architecture on translation across more than 50 languages. It activates 25 billion of its 218 billion parameters. Translation takes priority over general chat, with context and maximum output limited to 16K tokens each.

The published weights use CC BY-NC 4.0 for non-commercial use. Cohere also offers a Chat V2 API, free up to its applicable rate limit and governed by separate commercial agreements.

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 26 published benchmark results

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

instTrans_minor (Show measurement, test conditions, and source)

Source value: 26.89
Score: 26.89
Metric: off-target
Unit: %
Direction: Lower is better
Source type: vendor-reported
Evaluator: Bilibili Index-Translate evaluation
Status: active
Retrieved at: 2026-10-23
Methodology: Vendor-reported result. The Index-Translate low-resource table reports the North-Small-Translate (218B-A25B) result for instTrans_minor off-target. The page says this suite contains 2,793 instruction-translation tasks and reports quality and IFscore using GPT-5.6-Sol judgments. The page does not report a per-row model revision or inference settings.
Context: The evaluation page includes Bilibili's own Index-Translate models in the comparison. Bilibili competes in translation models, so these results are vendor-reported.

Source Bilibili Index-Translate evaluation

FLORES_minor_pair (Show measurement, test conditions, and source)

Source value: 13.2
Score: 13.2
Metric: off-target
Unit: %
Direction: Lower is better
Source type: vendor-reported
Evaluator: Bilibili Index-Translate evaluation
Status: active
Retrieved at: 2026-10-23
Methodology: Vendor-reported result. The North-Small-Translate (218B-A25B) row reports a 13.2% FLORES_minor_pair off-target rate. The page defines off-target as outputs in a language other than the target language and says lower is better. FLORES_minor_pair covers 104,000 inputs across 1,040 translation directions among 62 languages; the page does not state a per-row sample count, model revision, or inference settings.
Context: The evaluation page includes Bilibili's own Index-Translate models in the comparison. Bilibili competes in translation models, so these results are vendor-reported.

Source Bilibili Index-Translate evaluation

WMT26 standard (Show measurement, test conditions, and source)

Source value: 83.6
Score: 83.6
Metric: WMT26 standard reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: cohere.com
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Cohere evaluation, GPT-5.6-Sol judge. Agentic is a distinct workflow and must not be assigned to the standard checkpoint result.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source cohere.com

WMT26 agentic (Show measurement, test conditions, and source)

Source value: 84.36
Score: 84.36
Metric: WMT26 agentic reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: cohere.com
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Cohere evaluation, GPT-5.6-Sol judge. Agentic is a distinct workflow and must not be assigned to the standard checkpoint result.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source cohere.com

standard EU (Show measurement, test conditions, and source)

Source value: 82.17
Score: 82.17
Metric: standard EU reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: cohere.com
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Cohere evaluation, GPT-5.6-Sol judge. Agentic is a distinct workflow and must not be assigned to the standard checkpoint result.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source cohere.com

standard non-EU Europe (Show measurement, test conditions, and source)

Source value: 81.24
Score: 81.24
Metric: standard non-EU Europe reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: cohere.com
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Cohere evaluation, GPT-5.6-Sol judge. Agentic is a distinct workflow and must not be assigned to the standard checkpoint result.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source cohere.com

standard South Asia (Show measurement, test conditions, and source)

Source value: 86.16
Score: 86.16
Metric: standard South Asia reported score
Unit: No unit provided
Category: source-specific
Direction: Higher is better
Source type: vendor-reported
Evaluator: cohere.com
Status: active
Retrieved at: 2026-10-04
Methodology: Vendor-reported result. Cohere evaluation, GPT-5.6-Sol judge. Agentic is a distinct workflow and must not be assigned to the standard checkpoint result.
Context: Source-specific observation; it is not a shared comparison cohort. The source table does not state a unit or scale for this numeric score.

Source cohere.com

Cohere long-context translation evaluation (Show measurement, test conditions, and source)

Source value: 48.9
Score: 48.9
Metric: score as published
Unit: No unit provided
Benchmark version: Unversioned Cohere long-context evaluation described on the North Small Translate launch page
Direction: Higher is better
Source type: vendor-reported
Evaluator: Cohere, Introducing North Small Translate
Status: active
Retrieved at: 2026-10-17
Methodology: Vendor-reported result. Cohere reports 48.9 for North Small Translate on its long-context translation evaluation. The page does not disclose benchmark version, metric definition, prompts, decoding setup, judge, denominator, or evaluated item count.
Context: This is an underspecified vendor-reported score, not the separate WMT26 results already present in the baseline. Its scale and test set are unknown, so it is not numerically comparable to those WMT26 values.

Source Cohere, Introducing North Small Translate

North Small Translate internal output throughput, low concurrency (Show measurement, test conditions, and source)

Source value: 112
Score: 112
Metric: Output Tokens per Second (TOPS), source label; low-concurrency result
Unit: tokens/s
Direction: Higher is better
Source type: vendor-reported
Evaluator: Cohere, Introducing North Small Translate
Status: active
Retrieved at: 2026-10-18
Methodology: Vendor-reported result. Vendor-reported internal throughput test for the exact Cohere snapshot label North-Small-Translate-1.0. Cohere compares it with Gemma 4 31B TP1 (1 GPU) under identical, otherwise unspecified hardware and concurrency conditions. The source does not state accelerator model, exact concurrency level, quantization/route, runtime, tokenization or decoding settings, repetition count, or evaluation date.
Context: Cohere-reported internal test, not an independent evaluator run. Exact hardware and concurrency values are omitted from the text; low/high concurrency are the source's labels. No N or uncertainty is reported.

Source Cohere, Introducing North Small Translate

North Small Translate internal output throughput, high concurrency (Show measurement, test conditions, and source)

Source value: 39
Score: 39
Metric: Output Tokens per Second (TOPS), source label; high-concurrency result
Unit: tokens/s
Direction: Higher is better
Source type: vendor-reported
Evaluator: Cohere, Introducing North Small Translate
Status: active
Retrieved at: 2026-10-18
Methodology: Vendor-reported result. Vendor-reported internal throughput test for the exact Cohere snapshot label North-Small-Translate-1.0. Cohere compares it with Gemma 4 31B TP1 (1 GPU) under identical, otherwise unspecified hardware and concurrency conditions. The source does not state accelerator model, exact concurrency level, quantization/route, runtime, tokenization or decoding settings, repetition count, or evaluation date.
Context: Cohere-reported internal test, not an independent evaluator run. Exact hardware and concurrency values are omitted from the text; low/high concurrency are the source's labels. No N or uncertainty is reported.

Source Cohere, Introducing North Small Translate

Index-NativeLong, GuoFeng (zh→en) document-SEGALE/COMET (Show measurement, test conditions, and source)

Source value: 0.616
Score: 0.616
Metric: document-SEGALE/COMET
Unit: No unit provided
Benchmark version: Index-NativeLong
Category: translation
Direction: No direction published
Source type: independent-evaluator
Evaluator: Bilibili Index-Translate evaluation
Status: active
Retrieved at: 2026-10-19
Methodology: Independent evaluation. Evaluated on 274 document windows across GuoFeng, BWB Track A3, and Books; means equally weight five 4K–64K length bins; metric is document-SEGALE/COMET.
Context: The source gives 274 windows across all three datasets, not a per-column sample count; this table does not state a unit or score direction. The evaluation document is maintained by Bilibili, not Cohere.

Source Bilibili Index-Translate evaluation

Index-NativeLong, BWB Track A3 (zh→en) document-SEGALE/COMET (Show measurement, test conditions, and source)

Source value: 0.6282
Score: 0.6282
Metric: document-SEGALE/COMET
Unit: No unit provided
Benchmark version: Index-NativeLong
Category: translation
Direction: No direction published
Source type: independent-evaluator
Evaluator: Bilibili Index-Translate evaluation
Status: active
Retrieved at: 2026-10-19
Methodology: Independent evaluation. Evaluated on 274 document windows across GuoFeng, BWB Track A3, and Books; means equally weight five 4K–64K length bins; metric is document-SEGALE/COMET.
Context: The source gives 274 windows across all three datasets, not a per-column sample count; this table does not state a unit or score direction. The evaluation document is maintained by Bilibili, not Cohere.

Source Bilibili Index-Translate evaluation

Page 1 of 3

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

The 100 most popular models of this category, 0 to 100, higher is more popular

20 of 101 models

GPT-573

DeepSeek V4.1 Flash 101768

Kimi K368

DiffusionGemma 26B A4B IT63

Claude Opus 562

Qwen 3.8 Flash62

North-Small-Translate-1.00

DeepSeek V4.1 Flash 1017

DiffusionGemma 26B A4B IT

Claude Opus 5

Qwen 3.8 Flash

North-Small-Translate-1.0

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 5, 2026

North-Small-Translate-1.0Agnes 3.0 FlashAion 3.5Aion 3.5 MiniEmber-1

All specifications

Architecture

Model class: Mixture-of-experts model for machine translation; Source: Model class
Parameters: 218 billion total, 25 billion active; Source: Parameters

Input and output

Context window: 16K tokens according to Cohere; Source: Context window
Maximum output: 16K tokens according to Cohere; Source: Maximum output
Modalities: Text input and text output; Source: Modalities
Languages: More than 50 languages; Source: Languages

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking: Unknown; no source
Watermarking technique: Not known; no source

Licence and access

Access: API and downloadable weights; Source: Access
License: CC BY-NC 4.0; Source: License

API

API model ID: north-small-translate-1-0; Source: API model ID
Model ID: CohereLabs/North-Small-Translate-1.0; Source: Model ID

Further details

API interface: Cohere Chat V2; Source: API interface

Prices

API price: Free for trial and production keys up to the applicable rate limit; commercial agreement terms applySource

Frequently asked questions about North-Small-Translate-1.0

Can I run North-Small-Translate-1.0 locally?

Yes, the weights are open. With 218B parameters it needs a lot of memory, more a server than a laptop.

Stated by Cohere

Open weights: yes
License: CC BY-NC 4.0
Parameters: 218B
active per token: 25B

Sources and data date

Show all 10 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

Type	Evidence and data date
Research status	Research date October 4, 2026. 2 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Additional source	Cohere North Small Translate announcement (retrieved October 3, 2026; October 4, 2026; October 17, 2026; October 18, 2026) · Cohere North Small Translate announcement · Editorial description reviewed October 4, 2026
Additional source	Cohere API changelog (retrieved October 3, 2026) · Cohere API changelog · Editorial description reviewed October 4, 2026
Additional source	Cohere North Small Translate documentation (retrieved October 3, 2026) · Cohere North Small Translate documentation · Editorial description reviewed October 4, 2026
Additional source	Bilibili Index-Translate evaluation (retrieved October 19, 2026; October 23, 2026)
Additional source	OpenRouter: Popularity index (derived), CC BY 4.0
Additional source	Vercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional source	Ollama: cumulative downloads for the popularity index
Additional source	Replicate: cumulative runs for the popularity index
Additional source	Google Trending Now: limited feed selection for Germany and the US
Additional source	Creative Commons Attribution 4.0