Which hyperscaler is cheapest for OpenAI models in 2026?

Azure AI Foundry is the only hyperscaler with first-party access to OpenAI's frontier models including GPT-6 Astra, GPT-5, and GPT-5.6 Sol. AWS Bedrock has access to OpenAI gpt-oss open-weight models but not the closed GPT-5 family. Google Vertex AI has no first-party OpenAI access. For OpenAI workloads, Azure is the only choice; the comparison question becomes whether to use Azure PAYG (pay-as-you-go) or Azure PTU (Provisioned Throughput Units) at 30-50% off PAYG for predictable workloads exceeding 1 million daily tokens (Azure pricing page, September 2026).

Which hyperscaler is cheapest for Anthropic Claude in 2026?

Both AWS Bedrock and Google Vertex AI host Claude at the same per-token rates that Anthropic publishes directly. Bedrock is generally 15-25% cheaper than Azure for 10-50M token/month workloads running on Claude Sonnet, because Bedrock has no base platform fee and no provisioned-capacity minimum. For Claude Opus 4.7 specifically, Bedrock and Vertex list the same $5 input / $25 output per million tokens, with no surcharge on either platform (usage.ai, August 2026).

Which hyperscaler is cheapest for Google Gemini in 2026?

Google Vertex AI is the only hyperscaler with first-party access to Gemini models including 2.5 Pro, 2.5 Flash, and 3.0 Pro. Vertex AI pricing is 10-20% cheaper than equivalent Azure OpenAI models for Gemini-equivalent workloads. The free tier on Vertex AI ($300/90 days for new accounts) makes development significantly cheaper than Azure, which has no standing free tier. For Gemini, Vertex AI is the default choice — Bedrock and Azure have no equivalent (digiusher.com, March 2026).

What are Azure OpenAI PTUs?

Azure OpenAI Provisioned Throughput Units (PTUs) are reserved capacity commitments that provide guaranteed throughput and consistent latency for production SLAs. PTUs deliver 30-50% cost savings versus PAYG for predictable, high-volume workloads exceeding 1 million daily tokens with a minimum commitment. PTU pricing is roughly $1 per PTU per hour for GPT-4o class models. PTU is best for sustained production workloads; PAYG is best for development or bursty traffic (Azure pricing docs, September 2026).

What are the hidden costs on each platform?

All three platforms have hidden costs that can add 10-20% to the headline token bill. Azure OpenAI: PTU overcommitment, prompt inefficiency invisible to native tooling, supporting service charges (Monitor, Storage, networking), no hard budget enforcement. AWS Bedrock: Knowledge Bases on OpenSearch Serverless start at $345/month, Guardrails $0.15 per 1,000 text units, Flows $0.035 per 1,000 node transitions. Google Vertex AI: data pipeline charges, BigQuery storage $6.25/TB/month, cross-region egress adding 10-20%, agent runtime billed by the second (technologymatch.com, June 2026).

Does AWS Compute Savings Plan discount Bedrock inference?

No. AWS Compute Savings Plans apply to eligible EC2, Fargate, and Lambda usage only. Bedrock model-inference token charges are not discounted by Compute Savings Plans. AWS describes Compute Savings Plans as applying to compute services; Bedrock is metered and billed separately. For Bedrock savings, look at Provisioned Throughput reservations (1-month or 6-month commitments, 20-30% savings) or third-party tools like Holori that aggregate Marketplace pricing (usage.ai, August 2026).

How do model catalogs differ across the three hyperscalers?

AWS Bedrock hosts 100+ models from Anthropic, Meta, Mistral, Cohere, Stability, AI21, Amazon Titan, and Amazon Nova — but no closed OpenAI frontier models. Azure AI Foundry hosts 1,700+ models and has first-party exclusive access to OpenAI's GPT-5 family including GPT-6 Astra. Google Vertex AI hosts 200+ curated models and has first-party exclusive access to Gemini 2.5 Pro, Gemini 3.0 Pro, and Gemma 4. For multi-model flexibility across vendors, Bedrock is the broadest catalog; for OpenAI or Gemini, the choice is the matching hyperscaler (tech-insider.org, July 2026).

Azure OpenAI vs Bedrock vs Vertex Pricing 2026

Azure OpenAI vs Bedrock vs Vertex AI pricing 2026: Foundry exclusive on OpenAI GPT-5 family (1,700+ models, cached inputs ~90% off, PTU 30-50% off PAYG); Bedrock 100+ models, no base fee, 15-25% cheaper than Azure for 10-50M tokens/mo; Vertex 200+ Gemini-exclusive models, batch 50% off. Hidden costs add 10-20% on every platform.

Last verified: Sep 16, 2026.

At a glance

Azure AI Foundry: 1,700+ models; OpenAI GPT-5 family exclusive; PTU 30-50% off PAYG
AWS Bedrock: 100+ models; Anthropic, Meta, Mistral, Cohere, Stability; no base fee
Google Vertex AI: 200+ models; Gemini 2.5 Pro and Gemma 4 exclusive; batch 50% off
Bedrock vs Azure: 15-25% cheaper for 10-50M token/mo workloads
Vertex vs Azure: 10-20% cheaper for Gemini-equivalent workloads
Azure hidden costs: 15-40% platform overhead from PTU overcommitment
Bedrock hidden costs: Knowledge Bases $345/mo, Guardrails $0.15/1K units, Flows $0.035/1K transitions
Vertex hidden costs: Data pipelines, BigQuery $6.25/TB/mo, cross-region egress 10-20%

Three platforms, three exclusive catalogs

The first question in any hyperscaler LLM comparison is which vendor's models you actually need — each platform has exclusive access to a different frontier family.

Azure AI Foundry has first-party exclusive access to OpenAI's GPT-5 family including GPT-6 Astra, GPT-5, and GPT-5.6 Sol. The Foundry catalog lists 1,700+ models including OpenAI, Anthropic (via limited Azure-hosted routes), Meta, Mistral, and other open-weight families. Bedrock has first-party access to Anthropic Claude, Meta Llama, Mistral, Cohere, Stability, AI21, Amazon Titan, and Amazon Nova — but no closed OpenAI frontier models (only OpenAI gpt-oss open-weights). Vertex AI has first-party access to Gemini 2.5 Pro, Gemini 3.0 Pro, and Gemma 4 (tech-insider.org, July 2026).

For an OpenAI-or-nothing workload, Azure is the only option. For Anthropic, Bedrock is the natural choice. For Gemini, Vertex AI is the only choice. For multi-model flexibility across vendors, Bedrock is the broadest catalog. For hybrid workloads, the cost comparison must include the cross-region egress and Marketplace normalisation that each platform adds (usage.ai, August 2026).

Azure PTU vs PAYG: when reserved capacity pays back

Azure PTUs (Provisioned Throughput Units) deliver 30-50% cost savings versus PAYG for predictable workloads exceeding 1 million daily tokens — but the savings require a minimum commitment that does not work for bursty traffic.

PTUs are reserved hourly capacity with a guaranteed SLA on throughput and latency. Pricing is roughly $1 per PTU per hour for GPT-4o-class models — one PTU provides a fixed amount of inference capacity that translates to a specific token throughput. For a sustained workload of 10M input + 2M output tokens per day, PAYG pricing at $2.50/$10 per million tokens works out to roughly $50/day; PTU reservation of equivalent capacity runs at $30-$35/day, a 30-50% saving (digiusher.com, March 2026).

PTUs require a monthly or annual reservation commitment. PTUs not used are billed anyway. For bursty traffic, the PAYG model wins because the PTU capacity sits idle during low-traffic periods. The breakeven depends on workload predictability: workloads above 60-70% average utilisation benefit from PTU; workloads below that are cheaper on PAYG. Azure also supports Data Zone deployments and Global deployments at slightly higher rates (Azure pricing docs, September 2026).

Bedrock's add-on stack: the hidden cost

Bedrock inference is rarely the surprise line item — the add-on services stack that production deployments need often runs 20-35% of the inference cost.

Knowledge Bases (Bedrock's RAG service) runs on Amazon OpenSearch Serverless, which starts at $345 per month before a single query. Bedrock Guardrails cost $0.15 per 1,000 text units processed. Bedrock Flows (agent orchestration) bill $0.035 per 1,000 node transitions. Cross-region inference does not add a separate routing charge but has data-residency implications (technologymatch.com, June 2026).

On a production bill, these adjacent charges often run 20-35% of inference cost. Agentic workflows that chain multiple model calls (reasoning + tool selection + tool results + synthesis) can burn 5-10x the visible prompt-and-response tokens once intermediate reasoning is counted. Bedrock's value at that scale is governance, not discount — the per-token rates match Anthropic's direct pricing exactly, so the platform fee is what you pay for IAM integration and unified billing (technologymatch.com, June 2026).

Vertex AI's batch and free tier

Vertex AI's batch prediction 50% discount makes it the cost-competitive option for non-real-time inference workloads — embeddings generation, document summarisation, nightly report generation, and offline data processing.

Vertex AI Batch Mode cuts token rates by 50% across Gemini 2.5 Pro ($0.625/$5 per million tokens) and Gemini 2.5 Flash ($0.15/$1.25 per million tokens). Batch returns results within 24 hours and is the cheapest Gemini route. Vertex AI also has a free tier ($300/90 days for new Google Cloud accounts) that Azure does not match, making development significantly cheaper on Vertex (digiusher.com, March 2026).

Vertex AI's hidden costs come from data pipelines, BigQuery storage ($6.25 per TB/month after the first 1 TB free), Dataflow jobs ($0.018-$0.035 per vCPU hour), and Agent Engine (billed per second of agent runtime). Hybrid pricing complexity adds another layer: Vertex's pricing is separated into model pricing, generative AI pricing, and quota/throughput behaviour, requiring external normalisation for accurate cost-per-workload calculation. Cross-region egress adds 10-20% (usage.ai, August 2026).

What enterprise buyers should do next

Three actions for organisations choosing a hyperscaler LLM platform in 2026.

Run on PAYG for 30-60 days first. All three platforms look inexpensive at the model layer. Use PAYG pricing across your production workload before committing to PTU or Provisioned Throughput. Measure P95 throughput against reserved capacity before committing.
Model the add-on stack, not just the token rate. Bedrock Guardrails and Flows, Azure Monitor and AI Search, Vertex BigQuery and Dataflow — these adjacent services often run 15-40% of inference cost. Include them in your TCO model before signing.
Apply FOCUS normalisation across the three billing models. Azure bills on tokens and PTU capacity; Bedrock on tokens and request dimensions; Vertex on tokens plus compute plus data processing. Map all three to the FOCUS standard before drawing a price conclusion.

What to watch next

Three near-term datapoints. First, Azure OpenAI's introduction of native budget enforcement — Microsoft currently recommends alerts plus automation for stronger cost control, rather than hard limits. Second, AWS Bedrock's expansion of Custom Model Import beyond $0.0785/min/CMU pricing — the current per-CMU pricing is not directly comparable to token rates on other platforms. Third, Vertex AI's continued investment in adapter-based fine-tuning — the most flexible fine-tuning platform of the three, but the per-token training cost is not always published upfront.

#azure openai pricing 2026#aws bedrock pricing#vertex ai pricing#hyperscaler llm cost#ptu azure openai#bedrock guardrails

Written by

Fazlur RahmanAI, Entrepreneurship, EdTech

Fazlur Rahman is the founder of Tutorsbot, building AI-powered tools for learning and career growth. He writes about applying AI in real products and the practi… Read more

Fazlur Rahman is the founder of Tutorsbot, building AI-powered tools for learning and career growth. He writes about applying AI in real products and the practical side of building an ed-tech startup.