CIOPages

CIOPages · State of Play · Generative AI Platforms

State of Play 2026: Generative AI Platforms

Frontier models reach buyers directly from the labs and through the major clouds, and open-weight models can be self-hosted, on different prices and data terms. Where the market stands in 2026, and what tokens, reserved capacity and self-hosting cost where prices are public.

Published November 16, 2026

A generative AI platform gives applications access to large language models and other generative models through an API. The platform hosts and scales those models. The labs and cloud platforms bill for them per token of text in and out. This edition covers model access from the labs, the cloud platforms and self-hosted open-weight models. Agent runtimes are covered in AI Agent Orchestration.

Where the market stands

On November 18, 2025, Anthropic, Microsoft and NVIDIA announced that Claude models would be offered in Microsoft Foundry. Claude is also available on Amazon Bedrock and Google Cloud.

On April 22, 2026, Vertex AI became part of Google's Gemini Enterprise Agent Platform.

The EU AI Act's rules for general-purpose AI models have applied since August 2, 2025. Providers must give documentation of a model's capabilities and limitations to companies that build it into their own AI systems, and publish a summary of the content used to train it. Models released under a free and open-source license with public weights are exempt from the documentation duties, but not the training summary, unless they carry systemic risk. Models placed on the market before August 2, 2025 have until August 2, 2027 to comply. The Digital Omnibus on AI, Regulation (EU) 2026/1744, did not change these obligations.

On October 14, 2026, OpenAI notified developers that named versions of GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini leave its API on January 14, 2027.

Three ways to reach a model

  • Direct from the lab. The OpenAI API, the Claude API and the Gemini Developer API sell each lab's own models.
  • Through a cloud platform. Amazon Bedrock lists models from Anthropic, Meta, Mistral AI, Cohere, DeepSeek, xAI and others, plus OpenAI's open-weight gpt-oss models and Google's Gemma. Microsoft Foundry offers Models sold by Azure, which include Azure OpenAI models, alongside Claude. Gemini Enterprise Agent Platform offers Gemini models and partner models, including Claude.
  • Self-hosted open weights. OpenAI releases gpt-oss under the Apache 2.0 license. gpt-oss-120b fits on a single 80GB GPU, and gpt-oss-20b runs within 16GB of memory. Mistral releases Mistral Large 3 and Mistral Small 4 under Apache 2.0. Meta's Llama 4 Community License requires a company whose products had more than 700 million monthly active users on the Llama 4 release date to request a license from Meta. Anyone who distributes Llama 4, or a product that contains it, must display "Built with Llama" prominently.

One model can have more than one operator. On Amazon Bedrock, AWS operates inference for Claude. On Claude Platform on AWS, Anthropic operates it, and AWS provides sign-in, access control and billing. Anthropic directs organizations that need FedRAMP High, IL4, IL5 or HIPAA-ready compliance, or AWS as sole data processor, to Bedrock. In Microsoft Foundry, Claude runs as an Anthropic-operated service, on Azure or on Anthropic infrastructure. Microsoft states that Anthropic acts as an independent data processor there and that use is subject to Anthropic's terms.

What a token costs

List prices per million input and output tokens, as published on October 17, 2026.

  • OpenAI. gpt-6-astra costs $10 and $50, gpt-6.1-sol $2 and $10, and gpt-6-luna $0.10 and $0.50. Batch and Flex processing cost half. Fast mode, the former Priority processing, costs double.
  • Anthropic. Claude Fable 5.1 costs $10 and $50, Opus 5.5 $4 and $20, and Sonnet 5.5 $2 and $10. Haiku 5.5 costs $0.10 and $0.50 for prompts up to 100,000 tokens and $0.50 and $2.50 above that. The Batch API takes 50% off. Reading from the prompt cache costs 10% of the input price, 5% on Opus 5.5 and Sonnet 5.5, and 2.5% on Fable 5.1. Claude 4.7 and later models use a tokenizer that produces about 30% more tokens for the same text than earlier models.
  • Google. On Gemini Enterprise Agent Platform, Gemini 4 Argon costs $4 and $20 from December 1, 2026, and Gemini 3.8 Flash $1.50 and $7.50 from January 1, 2027. Google lists lower promotional prices for both models before those dates. Batch pricing for Gemini 4 Argon is half the standard rate.
  • Amazon Bedrock. Prices vary by model and region. Select models cost 50% less in batch. A Flex tier is discounted and a Priority tier carries a premium.

Keeping processing in one geography costs more. OpenAI adds 10% for regional processing on models released on or after March 5, 2026. Anthropic multiplies prices by 1.1 for US-only inference on Claude 4.6 and later. Anthropic says regional and multi-region endpoints for Claude 4.5 and later models cost 10% more than global ones on Bedrock and Google Cloud. Gemini 4 Argon and Gemini 3.8 Flash cost 10% more on non-global endpoints.

Reserved capacity

  • Bedrock Reserved tier. Terms of 1 or 3 months at a fixed price per 1,000 tokens per minute, with a minimum of 100,000 input and 10,000 output tokens per minute. Traffic above the reservation overflows to Standard. AWS targets 99.5% uptime. Billing continues until the reservation is deleted.
  • Microsoft Foundry provisioned deployments. Capacity is bought in provisioned throughput units (PTUs), with global, data zone or regional processing. Minimums vary by model. Batch costs 50% less than Global Standard, with a 24-hour target turnaround.
  • Google Provisioned Throughput. Sold in generative AI scale units (GSUs) on commitments from 1 week to 1 year. A global GSU costs $7.14 an hour on a 1-week term and $2.74 an hour on a 1-year term.
  • NVIDIA NIM. NVIDIA's inference containers for self-hosting are free, but the NIM Certified Production Branch requires an NVIDIA AI Enterprise subscription. List prices as published: $4,500 per GPU for 1 year, $22,500 per GPU for a perpetual license with 5 years of support, or $1 per GPU-hour on cloud marketplaces plus instance costs.

What the data terms say

  • Training. OpenAI says API data is not used to train its models unless the customer opts in. Anthropic's Commercial Terms say it may not train models on Customer Content. AWS states that Bedrock inputs and outputs are not used to train Amazon Nova, Amazon Titan or third-party models, and are not shared with model providers. Microsoft states that prompts and completions sent to Models sold by Azure are not used to train foundation models without the customer's permission or instruction. Google states it will not use customer data to train or fine-tune models without prior permission or instruction. On the free tier of the Gemini Developer API, content is used to improve Google's products.
  • Retention. OpenAI keeps abuse monitoring logs for up to 30 days, and zero data retention needs OpenAI's approval. Anthropic deletes API inputs and outputs within 30 days, with exceptions, unless agreed otherwise. Google caches Gemini data for up to 24 hours by default, which customers can turn off. Grounding with Google Search stores data for up to 3 days, and that cannot be turned off. Google may log prompts to detect abuse, and customers can request an exception. Microsoft may store flagged prompts for human abuse review, and managed customers can apply for modified abuse monitoring.
  • Location. Foundry Global deployments may process data in any Azure region. Data Zone deployments keep processing in the US, EU or Asia Pacific zone. OpenAI offers regional processing in the United States, Europe and the UAE.

What a model change requires

OpenAI gives at least 6 months' notice before retiring a generally available model, at least 3 months for specialized variants, and as little as 2 weeks for preview models. Anthropic gives at least 60 days' notice for publicly released models. Bedrock and Google Cloud set their own retirement dates for Claude, and some Claude models retired from the Claude API remain on them.

Claude in Microsoft Foundry does not support the Message Batches API, so the batch discount is not available there.

Keep a fixed set of your own prompts with graded answers. Run every candidate model through it, and run it again before each migration. Compare cost per task, not price per token, since tokenizers differ.

Questions to settle first

Settle these before choosing a platform.

  • Which workloads need processing kept in one geography, and what uplift does that add?
  • Who should be the data processor for each workload: the lab, the cloud provider or your own team?
  • What volume justifies reserved capacity or self-hosting over paying per token?
  • Which models in production have announced retirement dates, and who owns the migration?
  • Do you have written training and retention commitments for every route, including free tiers developers use?
Sources. Anthropic, "Microsoft, NVIDIA and Anthropic announce strategic partnerships", November 18, 2025: https://www.anthropic.com/news/microsoft-nvidia-anthropic-announce-strategic-partnerships; Google Cloud, Gemini Enterprise Agent Platform release notes: https://docs.cloud.google.com/gemini-enterprise-agent-platform/release-notes; Official Journal of the European Union, Regulation (EU) 2024/1689 (Artificial Intelligence Act): https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=CELEX:32024R1689; Official Journal of the European Union, Regulation (EU) 2026/1744 (Digital Omnibus on AI): https://eur-lex.europa.eu/legal-content/EN/TXT/HTML/?uri=OJ:L_202601744; OpenAI, API deprecations: https://developers.openai.com/api/docs/deprecations; OpenAI, API pricing: https://developers.openai.com/api/docs/pricing; OpenAI, Your data: https://developers.openai.com/api/docs/guides/your-data; OpenAI, gpt-oss repository on GitHub: https://github.com/openai/gpt-oss; Anthropic, Pricing: https://platform.claude.com/docs/en/about-claude/pricing; Anthropic, Claude Platform on AWS: https://platform.claude.com/docs/en/build-with-claude/claude-platform-on-aws; Anthropic, Claude in Microsoft Foundry: https://platform.claude.com/docs/en/build-with-claude/claude-in-microsoft-foundry; Microsoft Learn, Data, privacy, and security for use of Anthropic Claude models in Microsoft Foundry: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/claude-models/data-privacy; Anthropic, Model deprecations: https://platform.claude.com/docs/en/about-claude/model-deprecations; Anthropic, Commercial Terms of Service, effective June 17, 2025: https://www.anthropic.com/legal/commercial-terms; Anthropic Privacy Center, "How long do you store my organization's data?": https://privacy.claude.com/en/articles/7996866-how-long-do-you-store-my-organization-s-data; Google Cloud, Gemini Enterprise Agent Platform generative AI pricing: https://cloud.google.com/gemini-enterprise-agent-platform/generative-ai/pricing; Google, Gemini Developer API pricing: https://ai.google.dev/gemini-api/docs/pricing; Google Cloud, Gemini Enterprise Agent Platform and zero data retention: https://docs.cloud.google.com/gemini-enterprise-agent-platform/resources/zero-data-retention; AWS, Amazon Bedrock pricing: https://aws.amazon.com/bedrock/pricing/; AWS, Amazon Bedrock User Guide, Service tiers for optimizing performance and cost: https://docs.aws.amazon.com/bedrock/latest/userguide/service-tiers-inference.html; AWS, Amazon Bedrock FAQs: https://aws.amazon.com/bedrock/faqs/; Microsoft Learn, Understanding deployment types in Microsoft Foundry Models: https://learn.microsoft.com/en-us/azure/foundry/foundry-models/concepts/deployment-types; Microsoft, Data, privacy, and security for Foundry Models sold by Azure in Microsoft Foundry: https://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy; Mistral AI, Models: https://docs.mistral.ai/models; Meta, Llama 4 Community License Agreement: https://dev.meta.ai/llama/llama4/license/; NVIDIA, NIM for LLM and VLM, NIM Offerings: https://docs.nvidia.com/nim/large-language-models/latest/about-nim-llm/nim-offerings.html; NVIDIA, AI Enterprise Licensing Guide, Pricing: https://docs.nvidia.com/ai-enterprise/planning-resource/licensing-guide/latest/pricing.html

Ask to contribute

Vendors in this market can ask to answer the six questions for State of Play 2026: Generative AI Platforms. Use an email address on your company’s own domain: answers come from the company itself, not from an agency. We email a link to confirm the request.

Keep reading

Buyer Guide
Generative AI & LLM Platforms
Calculator
AI ROI Calculator

Estimate net ROI and payback period for your generative AI or AI/ML initiative.

State of Play
LLM observability

The annual review of this market: what changed, with answers from the vendors in it.

State of Play 2026: MLOps