Post-Training
Built for your Agent.

Post-Training
Built for your Agent.

→ Start training

↗ Read Docs

freesolo.co
/Freesolo⌄/people-search⌄
T

Overview

Your post-training workspace

Environments0
Training runs0
Live models0

Train your first specialized model

Build an environment, run post-training, then deploy an OpenAI-compatible endpoint.

Recent activityStatus
people-searchWaiting
claude code

Describe what you need

Approve one clear quote

Get a model you own

Pricing that makes sense

No per-token meter, no GPU-hour roulette. Your agent gets one quote for the whole run.

QWEN3-8B | SFT + GRPO | ~30M TOKENS

TINKER

≈ $420

TOGETHER

≈ $450

FLASH

$42

Specialization beats the frontier.

A sub-10B model tuned on your data beats a frontier model on your task, cheap and fast enough to retrain on the fly.

90

80

70

60

50

TASK ACCURACY (%)

0

200

400

600

800

1,000

GRPO TRAINING STEPS

FRONTIER · 73%

BASE MODEL

56%

SFT · STEP 0

71%

GRPO · STEP 1,000

89%

TASK ACCURACY (%)

50

60

70

80

90

0

200

400

600

800

1,000

GRPO TRAINING STEPS

FRONTIER · 73%

BASE MODEL

56%

SFT · STEP 0

71%

GRPO · STEP 1,000

89%

TASK ACCURACY (%)

50

60

70

80

90

0

500

1,000

GRPO TRAINING STEPS

FRONTIER · 73%

BASE MODEL

56%

SFT · STEP 0

71%

GRPO · STEP 1,000

89%

Custom kernels for training at light speed.

We treat kernel engineering as a search problem. Each permutation is continuously optimized with an autoresearch loop to ensure maximum training throughput.

Your data. Your model. Your weights.

WORKFLOW STEP

RECORDED

Own & export weights

Every run returns downloadable weights in standard formats. Serve them anywhere.

ACCURACY 94%

PASS

Your data stays isolated

Encrypted in transit and at rest and never used to train anything but your model.

SEARCH RELEVANCE

+31%

Reproducible & durable

Pinned configs and seeds, checkpointed end to end so every run always finishes.

Your data. Your model. Your weights.

WORKFLOW STEP

RECORDED

Own & export weights

Every run returns downloadable weights in standard formats. Serve them anywhere.

ACCURACY 94%

PASS

Your data stays isolated

Encrypted in transit and at rest and never used to train anything but your model.

SEARCH RELEVANCE

+31%

Reproducible & durable

Pinned configs and seeds, checkpointed end to end so every run always finishes.