Encord review, pricing and verdict

Encord prints no price for any plan. It lists Starter, Team and Enterprise, with a Get started button on the first two and Contact sales on Enterprise, and data volume limits that rise by plan: Index up to 500k, 100m and 1bn+, Active up to 50k, 1m and 10m, in the units the page prints. DICOM and NIfTI, 3D, LiDAR and point cloud, geospatial, ECG, LLM evaluations, custom data, on-prem deployment, a solutions architect and a custom MSA and SLA are marked as add-ons, so a quote depends on the data types chosen.

The platform has four working parts. Annotate labels images, video, audio, documents, DICOM and NIfTI medical imaging, and 3D, LiDAR and point-cloud data, with model-assisted labeling from Segment Anything Model 2 and object tracking. Index searches and curates large datasets with embeddings, outlier and duplicate detection, and Active analyses label errors and model performance. Post-training alignment adds RLHF, rubric-based evaluation and pairwise comparison for models already in production. Around them sit Data agents, a Python SDK and the encord-agents repository, Merlin, an agentic layer that the docs label BETA and that works in the Encord app or through MCP, and a managed Data-as-a-Service line of vetted annotators and in-field data collection.

Encord fits ML and data teams building robots, autonomous vehicles, drones and medical-imaging models, and frontier-model teams that need alignment data. Its own customer stories claim that Forterra sped up model iteration by 50 percent and that UiPath grew a dataset tenfold while cutting its error rate fourfold; both are vendor claims. On 20 October 2026 it released EMG-1, an open dataset of more than 300 hours of muscle-signal recordings paired with first-person video, built with Wearable Devices and launched alongside Encord Data Labs, its new Physical AI research division.

Buyers should look at alternatives when they need a published price, a free way in or a narrow job. Superb AI sells labeling and curation to industrial and mobility teams, Surge AI supplies expert human data for RLHF, and Scale AI is the larger managed-data vendor; search results for Encord alternatives on 11 November 2026 return roundups that name Labelbox, SuperAnnotate and V7, and none of the three has a Hokai record yet. Several data types sit outside the base plans on its pricing page, including DICOM and NIfTI, 3D, LiDAR, geospatial, ECG and LLM evaluations; Merlin is not available on on-premises or air-gapped deployments, and the open-source Encord Active repository is archived. For a first-hand look, on 11 November 2026 GitHub showed the encord-agents repository pushed on 10 November, the encord-client-python SDK on 28 October, and encord-active archived with its last push on 23 May 2025. More tools in this category are on the data science and ML platforms list, and best data science and ML tools compares the wider field.

The 2026 release notes carry 48 dated entries through 9 November. Merlin became native in the Encord app on 13 August and gained a default sidebar on 9 October, point-cloud labeling got one-click segmentation and cuboids on 13 August, Task Rules and the Metadata Router reached general availability on 8 and 2 September, the Timeseries modality on 16 October and project-level Issue Management on 5 November.