Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors

We show this table for reference; we do not rank on it.

Cohere launch reranking report.

Evaluation results published by Cohere.

Expected calibration error on Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors — October 21, 2026 launch; November 4 capture

We mirror the published expected calibration error view for Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors. Cohere Rerank 5 Pro leads the public snapshot at 0.057, followed by Cohere Rerank 5 Fast (0.058) and Voyage AI Rerank 3 (0.307). We do not use these results to rank models overall.

Cohere Rerank 5 Pro

Cohere

Cohere provider report; evaluated snapshot and hardware undisclosed

0.057

Cohere Rerank 5 Fast

Cohere

Cohere provider report; evaluated snapshot and hardware undisclosed

0.058

Voyage AI Rerank 3

Voyage AI

Cohere provider report; evaluated snapshot and hardware undisclosed

0.307

5 configurationsCross-domainDated provider configurationsDisplay onlyUpdated October 21, 2026 launch; November 4 capture

Should you trust a Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors score?

Read with care: scores cluster at the top; the benchmark is not marked current.

Saturation Active: Top three within 0.250 score units; frontier rows are clustering, so read gaps at the top with care.
Who produced the rows Mirrored table: Every row is mirrored from Cohere Rerank 5 launch report as it publishes it; BenchLM did not re-run these results. Source.
Freshness and weight Display only: Dated provider configurations, refreshed reviewed dated source capture. Display only; it does not move rankings. Upstream datasets and terms; aggregate results only.

Row statuses come from the verification registry; saturation and freshness from the benchmark lifecycle. A provider self-report is displayable evidence, not an independent run.

Expected calibration error table (5 configurations)

Expected calibration error results for Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	Cohere Rerank 5 ProCohereCohere provider report; evaluated snapshot and hardware undisclosed	0.057	Not reported	Unknown
2	Cohere Rerank 5 FastCohereCohere provider report; evaluated snapshot and hardware undisclosed	0.058	Not reported	Unknown
3	Voyage AI Rerank 3Voyage AICohere provider report; evaluated snapshot and hardware undisclosed	0.307	Not reported	Unknown
4	Cohere Rerank 4 ProCohereCohere provider report; evaluated snapshot and hardware undisclosed	0.309	Not reported	Unknown
5	Voyage AI Rerank 3 LiteVoyage AICohere provider report; evaluated snapshot and hardware undisclosed	0.325	Not reported	Unknown

How to read this provider report

Expected calibration error of raw relevance scores against NIST TREC Deep Learning assessors: 20,646 documents, useful = grade 2 or higher on the 0-3 scale. Lower is better; 0 is perfect. Scores used as returned, with no recalibration.

Cohere reports these results in its Rerank 5 launch. The linked full spreadsheet requires sign-in at capture, so this view includes only exact figures available in the public post. Model snapshots, evaluation dates, candidate-set sizes, and uncertainty intervals are undisclosed.

Each metric retains its own cohort. Rerankers reorder candidates from a first-stage search; these scores do not measure embedding or keyword retrieval on their own. These provider comparisons do not enter weighted model rankings.

Snapshot

Cohere launch and charts Rerank API model details OpenRouter availability (November 4, 2026)Provider full-results sheet (sign-in required at capture)RCP-nDCG methodology and annotations

The published Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors snapshot places Cohere Rerank 5 Pro first at 0.057. The third row is 0.250 score units higher. The broader top-10 range is 0.268 score units, so many of the published results sit in a relatively narrow band.

5 configurations are shown for Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors. The benchmark falls in the Cross-domain category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors

Year

2026

Public benchmark source

Freshness and provenance

Version

Cohere Rerank 5 launch provider report

Refresh cadence

Reviewed dated source capture

Staleness state

Dated provider configurations

Question availability

Upstream datasets and terms; aggregate results only

Dated provider configurationsDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

Questions

What does Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors measure?

Cohere launch reranking report.

Which model leads the published Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors snapshot?

Cohere Rerank 5 Pro currently leads the published Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors snapshot with 0.057 expected calibration error. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Cohere Rerank 5 launch: calibration error vs NIST TREC-DL assessors?

The October 21, 2026 launch; November 4 capture snapshot contains 5 AI models.

Last updated: October 21, 2026 launch; November 4 capture · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.