Key facts
| AWS Bedrock API surface | OpenAI-compatible Chat Completions and Responses via the bedrock-mantle endpoint; legacy InvokeModel/Converse also supported |
| AWS Bedrock pricing model | Usage-based per-model billing through your AWS account; no flat self-serve plan |
| AWS Bedrock deployment | AWS-managed and region-bound; IAM, VPC endpoints and Guardrails built in |
| Plugsky API compatibility | OpenAI-compatible /v1/chat/completions; change the base URL and model name |
| Plugsky pricing model | Flat monthly self-serve plans with unlimited fair-use usage; no per-token billing |
| Plugsky model access | 30+ models behind one API, from free to frontier tiers |
| Plugsky deployment and residency | Plugsky cloud, your VPC, on-prem and air-gapped; region selection for residency |
| Plugsky migration effort | One line for OpenAI-style code: change base_url and model name |
TL;DR
- Bedrock is AWS-native; Plugsky is OpenAI-compatible — the migration paths differ but both are short.
- Plugsky self-serve pricing is flat monthly with unlimited fair-use usage, not per-token; the free plan starts with 2 free AI models and a 14-day full-access trial.
- 30+ models sit behind one Plugsky API, from free tiers to frontier reasoning.
- Plugsky adds VPC, on-prem and air-gapped deployment plus region selection.
- Where Bedrock still wins: AWS IAM and billing integration, Guardrails and AWS-exclusive model access.
How it works, step by step
- Inventory every Bedrock InvokeModel/Converse or bedrock-mantle call, model ID and region in your codebase.
- Create a Plugsky account and API key on the free plan (no card required).
- Map each Bedrock model ID to the closest Plugsky model and record the mapping.
- Wrap your model client behind an interface, then add a Plugsky client for the same prompts.
- Run your evals side by side: quality, latency, streaming, tool calls and JSON mode.
- Move one non-critical workload first and compare error rates for a week.
- Cut over the rest and keep the Bedrock adapter behind a feature flag for rollback.
Try it yourself
Open the LLM cost calculator →
Where AWS Bedrock wins
Bedrock is the natural choice when your platform is already AWS. Requests stay inside your AWS account boundary, cost allocation tags and CloudWatch metrics work as usual, and IAM, VPC endpoints, Guardrails and private pricing agreements all apply. The bedrock-mantle endpoint also exposes OpenAI Chat Completions, the Responses API and the Anthropic Messages API, so an OpenAI SDK app can often run on Bedrock with minimal changes. If you need Bedrock-exclusive models or AWS commitment spend, that advantage is real and hard to replace.
Where Plugsky wins
Plugsky is OpenAI-compatible at the primary endpoint, not as a compatibility layer, and it is cloud-agnostic: you do not need an AWS account, VPC endpoint or IAM policy to make a first call. Self-serve plans are flat monthly with unlimited fair-use usage, so high-volume chat and agent loops do not translate into token-meter anxiety. One key reaches 30+ models across families, and model routing, fusion and failover are built into the platform. Enterprise deployments add your VPC, on-prem and air-gapped options plus region selection for residency.
What to test before switching
Run the same prompt set through both stacks and compare: streaming SSE behaviour, function-calling and tool loops, JSON mode adherence, long prompts, embeddings and error semantics (429, 5xx, retry headers). Measure latency from the region your users actually sit in, not from a laptop. Confirm where request and response data is stored and logged. If you depend on Guardrails, map each policy to an application-side check before you migrate, and test Bedrock-only features separately so they do not block the chat migration.
Migration path from Bedrock to Plugsky
If your app calls the bedrock-mantle OpenAI-compatible route, the switch is a base URL, API key and model ID change — for example client = OpenAI(base_url='https://api.plugsky.com/v1'). If it calls legacy InvokeModel or Converse, add a thin translation layer that converts the Bedrock request shape to /v1/chat/completions and map stop reasons and tool-call fields back. Keep the translation layer behind your existing model interface so rollback is a config flag. Then follow the hybrid Bedrock workflow guide to run both side by side during cutover.
Honest comparison
| Capability | Plugsky | AWS Bedrock | Building in-house |
|---|---|---|---|
| API compatibility | OpenAI-compatible; base URL + model name | OpenAI-compatible via bedrock-mantle; legacy SDK also available | Full rewrite |
| Pricing model | Flat monthly self-serve, unlimited fair use | Usage-based per model via AWS | GPU + ops cost |
| Model access | 30+ models, one API and one key | Managed catalogue bound to AWS regions | You host every model |
| Deployment and residency | Cloud, VPC, on-prem, air-gapped; region choice | AWS regions only, inside your AWS account | You control |
| Migration effort | One line for OpenAI-style code | Minimal on bedrock-mantle; adapter for legacy SDK | Months |
| Fine-tuning and custom models | Not yet — fine-tuning is coming soon | Available for selected models | You own the pipeline |
Frequently asked questions
Can I point my OpenAI SDK at Plugsky instead of Bedrock?
Yes. Plugsky exposes an OpenAI-compatible chat completions endpoint, so you change the base URL and model name and keep your SDK code. Bedrock's mantle route works the same way, so this part is symmetric.
Bedrock already has an OpenAI-compatible endpoint — why switch?
The API shape is similar; the differences are commercial and operational. Plugsky offers flat monthly self-serve pricing, a free plan with two models, one key across 30+ models, and deployment options outside AWS regions.
How does Plugsky pricing differ from Bedrock?
Bedrock bills per model and token through your AWS account. Plugsky self-serve plans are flat monthly with unlimited fair-use usage and no per-token charges. See the live pricing page for current plans.
Does Plugsky support streaming and function calling?
Yes — streaming, function calling and JSON mode are live, as are embeddings, RAG and agents. Audio, images, batch and fine-tuning are coming soon.
Can Plugsky run inside our AWS VPC?
Yes. Plugsky supports VPC, on-prem and air-gapped deployments for enterprise customers, with region selection for data residency.
What happens to Bedrock-exclusive models?
Keep them on Bedrock. Migrate the OpenAI-style chat and reasoning workloads you can evaluate, and run a hybrid stack while you validate quality and reliability.
How much migration work is involved?
For OpenAI-style code paths it is a base URL and model mapping. For the legacy AWS SDK request shape, you add a translation layer and run your evals before cutover.
Can I keep Bedrock and Plugsky running side by side?
Yes. Route a share of traffic to Plugsky behind a feature flag, compare quality and reliability on real prompts, then expand or roll back.