2pm.spaceLLM fine-tuning platforms:
the alternatives, compared
Training is the short part of fine-tuning. Collecting good examples and putting the tuned model to work are what take the time.
What to look for
- Where the training data comes from, and how you review it before training.
- Which base models you can tune, and with which method (LoRA or full).
- Price per training token, and what the tuned model costs to serve.
- How the tuned model reaches production — an API endpoint, or a model your agents already pick from.
The 8 tools
- 1.
2pm.space
Our productFine-tuning built into the inbox: rate and correct your agent's replies, train a LoRA adapter on open-weight models such as Llama, Qwen and Mistral, then pick it from any agent's model list — billed from the same credit.
- Best for
- Teams tuning a support or sales agent on real conversations
- Pricing
- Free for every feature — you pay only for the AI you use, at the model provider's own rate
- 2.
Together AI
Cloud service for supervised and preference (DPO) fine-tuning of open-weight models such as Llama, Qwen, Gemma and DeepSeek, billed per token processed, with deployment on Together's inference platform.
- Best for
- Developers fine-tuning and hosting open-weight models via API
- Pricing
- From $0.34 per million training tokens
- 3.
Fireworks AI
Managed fine-tuning (LoRA and full-parameter SFT/DPO) of open models on Fireworks' inference cloud, plus per-GPU-hour training; fine-tuned models serve at base-model prices.
- Best for
- Teams fine-tuning open models and serving them at scale
- Pricing
- From $0.50 per million training tokens
- 4.
Amazon Bedrock
AWS service for supervised and reinforcement fine-tuning and continued pre-training of Bedrock models such as Amazon Nova, Titan, Llama, Cohere, gpt-oss and Qwen, billed per training token or hour.
- Best for
- AWS teams customizing foundation models inside their AWS account
- Pricing
- From $0.0004 per thousand training tokens (Titan Text Lite)
- 5.
Google Gemini Enterprise Agent Platform (Vertex AI)
Google Cloud's supervised, preference and reinforcement fine-tuning for Gemini models plus open models such as Gemma, Llama and Qwen, now part of Gemini Enterprise Agent Platform (formerly Vertex AI).
- Best for
- Google Cloud teams tuning Gemini or open models
- Pricing
- From $0.28 per million training tokens (Llama 3.2 1B)
- 6.
Replicate
API platform for running, fine-tuning and deploying open-source models, billed per second of hardware; fine-tuning is centred on image models such as FLUX. Replicate joined Cloudflare in November 2025.
- Best for
- Developers fine-tuning image models like FLUX via API
- Pricing
- From $0.000225 per GPU-second (Nvidia T4)
- 7.
Unsloth
Open sourceFree planOpen-source framework (Apache-2.0 core, AGPL-3.0 Studio UI) for faster, lower-memory LLM fine-tuning and local inference, with Pro and Enterprise tiers priced on request.
- Best for
- Developers fine-tuning open models on their own GPUs
- Pricing
- Open source, free to self-host
- 8.
Axolotl
Open sourceFree planFree, Apache-2.0 open-source framework for fine-tuning and post-training LLMs (LoRA, QLoRA, full fine-tuning, preference and RL methods) via YAML configs, run on your own or rented GPUs.
- Best for
- ML engineers fine-tuning open models on their own infrastructure
- Pricing
- Open source, free to self-host
Frequently asked questions
Hosted models like Claude cannot be LoRA-tuned by customers. Fine-tuning on 2pm.space uses open-weight base models such as Llama, Qwen, Mistral and gpt-oss.
On 2pm.space, from your inbox: replies you rate up, and corrections you write with Improve. Every pair waits for your Include before it is exported.
Not on 2pm.space — a dataset trains an adapter that belongs to your workspace only. The training itself runs at Together AI, with phone numbers and emails scrubbed first.
How we checked
Every description on this page comes from the vendor's own public website, checked on October 4, 2026. Products and prices change, so check each vendor's site for the current version — and if something here is out of date, tell us and we will correct it.
Prices are a starting list price each vendor shows for a paid plan, before tax and add-ons; some are for annual billing. Check the vendor's pricing page for the plan you need.
Product names and trademarks belong to their owners. 2pm.space is not affiliated with, or endorsed by, the companies named here.
More comparisons
Try 2pm.space — every feature is free
No plans and no seats. Open a workspace, connect a channel or a database, and pay only for the AI your agents use.