Vendor notes

The widest open-model menu on our list, from sub-1B Qwen3.5 models up to Qwen3.5 397B A17B, DeepSeek and GLM-5.x. Its pricing page shows SFT at $0.34 per million tokens for Llama 3.1 8B and $2.03 for Llama 3.3 70B, with DPO at $0.84 and $5.08.

Billed tokens are epochs times training tokens plus evaluations times validation tokens, so frequent validation adds cost. Every job carries a minimum charge: $4.00 for most models, rising to $60.00 for the largest. Its docs say failed jobs are fully refunded.

From Pricing, Together AI and Fine-tuning pricing, Together AI docs, as viewed on 7 Nov 2026.

Google's Gemini models and open models (Gemma, Llama, Qwen) are tuned from the same price page. Methods vary by model: supervised tuning across the list, reinforcement learning fine-tuning on Gemini 3.5 Flash, and preference tuning on Gemini 2.5 Flash and Flash Lite.

Watch the units. Gemini rows are priced per 1,000 training tokens (Gemini 3.5 Flash at $0.01), open models per million (Llama 3.1 8B at $0.67, Llama 3.3 70B at $6.72). Training tokens are dataset tokens times epochs.

From Generative AI on Vertex AI pricing, Google Cloud, as viewed on 7 Nov 2026.

The clearest price sheet of the eight: managed training is priced by base-model size, with LoRA SFT from $0.50 per million training tokens (up to 16B) to $10.00 (over 300B), and full SFT at double the LoRA rate in each tier.

Dedicated training on rented GPUs starts at $8.00 per GPU-hour (H100 or H200). Serving a tuned model adds nothing on top, Fireworks says: the bill is the training job plus deployment time, counted in GPU-seconds.

From Pricing, Fireworks AI, as viewed on 7 Nov 2026.

Reinforcement fine-tuning of OpenAI gpt-oss-20b or Qwen3 32B is priced at $80.00 per training hour, and the tuned models are then served on demand per token. Older Llama 2 fine-tuning is priced per million training tokens ($1.49 for 13B, $7.99 for 70B) and served on provisioned throughput at $23.50 per model unit per hour without commitment.

Every custom model carries a storage fee of $1.95 per month. Our capture of this page was partial (US East and US West prices); tables for Amazon Nova and Titan, Cohere, Llama 3.x and distillation were cut off, so read those on the page itself.

From Bedrock pricing, AWS, as viewed on 7 Nov 2026.

The Azure OpenAI price list we read covers OpenAI models only: reinforcement fine-tuning for o4-mini (per training hour plus hosting per hour) and supervised tuning for the GPT-4.1 and GPT-4o families (per million training tokens plus hosting per hour).

The pricing page renders its figures in the browser, so we could not record them. Every fine-tunable model there carries a hosting line billed per hour as well as a training price, so model both before comparing it with per-token serving elsewhere.

From Azure OpenAI Service pricing, Microsoft, as viewed on 7 Nov 2026.

A Ray-based compute platform rather than a fine-tuning API. According to its docs, post-training runs on Ray Train, SkyRL or LLaMA-Factory, and can be supervised, DPO or reinforcement learning with verifiable rewards (RLVR), as LoRA adapters or full-weight updates.

Suited to teams that want to run their own training code on managed clusters. Pricing is by quote, so the cost comparison has to come from a conversation.

From LLM fine-tuning, Anyscale docs, as viewed on 7 Nov 2026.

Prime Intellect launched its Environments Hub on 27 August 2025: an open platform for building and sharing reinforcement learning and evaluation environments. It sits closer to the reinforcement fine-tuning end of the market than to supervised tuning.

Relevant if your project depends on a grader or environment rather than a file of examples. No prices were published on the pages we read.

From Environments Hub announcement, Prime Intellect blog, as viewed on 7 Nov 2026.

CoreWeave announced on 3 September 2025 that it was acquiring OpenPipe, which it described as a leader in reinforcement learning.

We have not yet read OpenPipe's own product or pricing pages, which is why every square is empty. That is a gap in our reading, not a finding about the product. It will be scored in the next check.

From OpenPipe acquisition news, CoreWeave, as viewed on 7 Nov 2026.

According to OpenAI's own API pricing page, new users can no longer sign up for fine-tuning. Teams already on it may keep creating training jobs "for the coming months"; their tuned models remain callable until each base model is deprecated. A new project cannot start there, so it carries no score.

From API pricing, OpenAI, as viewed on 7 Nov 2026.