# Ray vs Together AI

**Leader by Vioscale score:** Together AI

| Attribute | Ray | Together AI |
|---|---|---|
| **Vioscale score** | 66.8 (68% (medium)) | 72.8 (60% (medium)) |
| activity.commits_last_30d | 100 | - |
| adoption.dependent_repos | 3,641 | - |
| adoption.github_stars | 43,612 | - |
| adoption.package_downloads_weekly | 11,291,327 | - |
| deployment.options | `{"cloud":true,"hybrid":true,"on_prem":true,"self_hosted":true}` | `{"cloud":true}` |
| description.long | Ray is an open-source unified framework for scaling Python and machine learning applications across any infrastructure. It provides distributed compute primitives, specialized AI libraries for data processing, model training, tuning, and serving, with seamless scaling from development environments to large clusters. | Together AI provides cloud infrastructure for hosting and serving open-source language, image, audio, and video models through serverless inference, reserved capacity, and dedicated GPU instances. The platform includes fine-tuning capabilities, batch processing, managed storage, and GPU cluster support for custom model development and training—eliminating the need for users to manage underlying infrastructure. |
| features.capabilities | `{"managed_cloud":true,"model_serving":true,"self_hostable":true,"multi_provider":true,"vpc_deployment":true,"framework_agnostic":true,"experiment_tracking":true}` | `{"role":"platform","evaluation":false,"soc2_type_ii":false,"compute_model":"multi_tenant_hybrid","managed_cloud":true,"model_serving":true,"pricing_model":"per_million_tokens","self_hostable":false,"multi_provider":true,"vpc_deployment":true,"otel_compatible":false,"no_train_on_data":"yes","llm_observability":false,"prompt_management":false,"framework_agnostic":true,"experiment_tracking":false,"openai_compatible_rest_api_schema":false,"bring_your_own_weights_byow_hosting":true,"dynamic_batching_and_kv_cache_management":false,"massive_context_window_support_100k_plus":false,"hipaa_baa_compliance_for_medical_inference":false,"multimodal_vision_and_audio_in_out_support":true,"native_function_calling_and_strict_json_mode":false,"streaming_server_sent_events_sse_token_yield":false,"lpu_or_custom_silicon_for_extreme_low_latency":false,"zero_data_retention_training_opt_out_enterprise":true}` |
| integrations.count | 17 | - |
| integrations.list | `[{"name":"PyTorch"},{"name":"TensorFlow"},{"name":"AIBrix"},{"name":"AReaL"},{"name":"Cosmos Curate"},{"name":"Daft"},{"name":"Data-Juicer"},{"name":"DeltaCAT"},{"name":"Modin"},{"name":"NeMo Curator"},{"name":"NeMo-RL"},{"name":"OpenRLHF"},{"name":"RayDP"},{"name":"ROLL"},{"name":"SkyRL"},{"name":"SLIME"},{"name":"Syftr"},{"name":"verl"},{"name":"vLLM"}]` | - |
| language.primary | Python | - |
| license.spdx | Apache-2.0 | - |
| market.availability | - | `{"primaryMarkets":["US"],"availabilityScope":"global","availableCountries":[],"notAvailableCountries":[]}` |
| platform.support | `{"cli":true}` | `{"cli":true,"web":true}` |
| pricing | `{"type":"free","freeTier":true,"sourceUrl":"https://www.ray.io","retrievedAt":"2026-08-14T16:39:27.110Z"}` | `{"type":"usage","plans":[{"free":false,"name":"Serverless Inference","summary":"From $0.00014–$15/1M tokens depending on model. Pay only for usage.","features":["50+ open-source models","Variable pricing by model and token type","Batch API support","Private endpoints"],"components":[{"per":{"qty":1000000,"unit":"tokens"},"kind":"metered","amount":0.00014,"currency":"USD"}],"contactSales":false},{"free":false,"name":"Provisioned Throughput","summary":"Reserved throughput capacity with 99% SLA. PTU-based pricing structure.","features":["Reserved token capacity","99% uptime SLA","Token-based pricing model","Production-grade reliability"],"commitment":"monthly","contactSales":false},{"free":false,"name":"Dedicated Inference","summary":"$3.69–$8.99 per GPU per hour (on-demand); reserved discounts available.","features":["Single-tenant GPU instances","Guaranteed performance (no resource sharing)","Custom model support","Autoscaling for traffic spikes"],"commitment":"monthly","components":[{"kind":"fixed","amount":3.69,"period":"hour","currency":"USD"}],"contactSales":false}],"addOns":[{"name":"Fine-Tuning","components":[{"per":{"qty":1000000,"unit":"tokens"},"kind":"metered","amount":0.48,"currency":"USD"}]},{"name":"Managed Storage"},{"name":"GPU Clusters"}],"summary":"Usage-based pricing starting from $0.00014/1M tokens for serverless inference. Reserved capacity and dedicated GPU instances available; on-demand GPU pricing from $3.69/hour.","currency":"USD","freeTier":false,"sourceUrl":"https://www.together.ai/pricing","retrievedAt":"2026-08-21T12:44:34.463Z","startingPrice":{"unit":"tokens","amount":0.00014,"currency":"USD"},"billingPeriods":["month"]}` |
| pricing.free_tier | yes | no |
| pricing.model | commercial | commercial |
| pricing.price_level | free | low |
| pricing.starting_price | - | `{"amount":0.00014,"currency":"USD"}` |
| pricing.transparent | - | yes |
| release.cadence_days | 18 | - |
| release.history | `[{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.58.0","date":"2026-08-23T05:42:08Z","type":"stable","version":"ray-2.58.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.57.0","date":"2026-08-11T01:15:44Z","type":"stable","version":"ray-2.57.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.56.1","date":"2026-07-17T23:19:47Z","type":"stable","version":"ray-2.56.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.56.0","date":"2026-06-29T20:32:55Z","type":"stable","version":"ray-2.56.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.55.1","date":"2026-04-22T20:24:41Z","type":"stable","version":"ray-2.55.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.55.0","date":"2026-04-15T20:34:27Z","type":"stable","version":"ray-2.55.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.54.1","date":"2026-03-25T23:37:39Z","type":"stable","version":"ray-2.54.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.54.0","date":"2026-02-18T23:44:44Z","type":"stable","version":"ray-2.54.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.53.0","date":"2025-12-20T15:16:24Z","type":"stable","version":"ray-2.53.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.51.2","date":"2025-11-29T00:40:40Z","type":"stable","version":"ray-2.51.2"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.52.1","date":"2025-11-28T02:23:22Z","type":"stable","version":"ray-2.52.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.52.0","date":"2025-11-21T19:10:39Z","type":"stable","version":"ray-2.52.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.51.1","date":"2025-11-01T03:27:11Z","type":"stable","version":"ray-2.51.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.51.0","date":"2025-10-29T05:33:49Z","type":"stable","version":"ray-2.51.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.50.1","date":"2025-10-18T19:21:43Z","type":"stable","version":"ray-2.50.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.50.0","date":"2025-10-10T23:06:29Z","type":"stable","version":"ray-2.50.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.49.2","date":"2025-09-19T18:10:36Z","type":"stable","version":"ray-2.49.2"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.49.1","date":"2025-09-03T00:44:11Z","type":"stable","version":"ray-2.49.1"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.49.0","date":"2025-08-26T19:52:24Z","type":"stable","version":"ray-2.49.0"},{"url":"https://github.com/ray-project/ray/releases/tag/ray-2.48.0","date":"2025-07-18T22:27:25Z","type":"stable","version":"ray-2.48.0"}]` | - |
| reliability.sla_pct | - | 99 |
| reliability.status_page | yes | yes |
| security.disclosure_policy | - | yes |
| security.gdpr | - | yes |
| security.iso27001 | - | yes |
| security.scorecard | 5.7 | - |
| security.soc2 | - | yes |
| security.vulnerabilities | `{"count":12,"source":"https://advisories.ecosyste.ms/api/v1/advisories?ecosystem=pypi&package_name=ray&per_page=100","last_12m":6,"max_severity":"CRITICAL"}` | - |

## Capabilities (MLOps & LLMOps Tools)

| Capability | Ray | Together AI |
|---|:--:|:--:|
| **Core** |  |  |
| Tool role | - | End-to-end ML platform |
| **Deployment** |  |  |
| Self-hostable / OSS core | ✓ | ✗ |
| Managed cloud available | ✓ | ✓ |
| On-prem / VPC deployment | ✓ | ✓ |
| **Observability** |  |  |
| LLM tracing / observability | - | ✗ |
| Evaluation (offline / LLM-judge / human) | - | ✗ |
| **Dev** |  |  |
| Prompt management + versioning | - | ✗ |
| **Tracking** |  |  |
| Experiment tracking / model registry | ✓ | ✗ |
| **Serving** |  |  |
| Model serving / inference endpoint | ✓ | ✓ |
| **Interop** |  |  |
| OpenTelemetry / OpenLLMetry compatible | - | ✗ |
| Framework-agnostic | ✓ | ✓ |
| **Gateway** |  |  |
| Multi-provider model support | ✓ | ✓ |
| **Data** |  |  |
| No-train-on-customer-data guarantee | - | Yes |

*Source: Vioscale. Generated 2026-09-01T15:09:27.494Z. "-" = undocumented, not absent.*