Castform Pricing 2026
Official pricing checked
Qwen3.5-4B rate card
USD per 1 million tokens; inference and training are separate meters.
| Usage | Model | Input / 1M | Output / 1M | Training / 1M |
|---|---|---|---|---|
| Hosted inference | Qwen3.5-4B | $0.03 | $0.15 | — |
| RL training | Qwen3.5-4B | — | — | $0.10 introductory |
What Castform charges for
Castform is a reinforcement-learning platform for post-training open models. Its public rate card separates hosted inference from training. The canonical pricing row above is loaded from our daily-updated dataset, which reads Castform's official public pricing API.
What the Neon benchmark changes
Castform and Neon report that a tuned Qwen3.5-4B retrieval agent scored 1.447 mean reward at $0.000920 per rollout, versus GPT-5.6 Sol at 1.369 reward and $0.087338. The displayed ratio is 94.9×, rounded to “100×” in the headline.
That is a vendor-reported workload result, not a general model ranking or a guaranteed discount. The public hosted rate does not identify the benchmark checkpoint, and total cost also includes training tokens, synthetic data, embeddings, Neon compute, evaluation, monitoring, and fallback calls. Read the full Castform versus GPT-5.6 Sol analysis.
Labs availability
The tuned retrieval specialist has an explicit Labs blocker. A valid test needs the exact callable artifact, fixed corpus and holdout, published reward weights and stopping rules, plus measured token and database usage. Running the public base endpoint would not test the post-trained result.
Buyer checklist
- Confirm whether your account uses the published introductory training rate.
- Measure cost per accepted answer, including retrieval and fallback infrastructure.
- Keep a blind holdout separate from synthetic training questions.
- Verify the deployed endpoint is the tuned artifact you evaluated.
Frequently asked questions
How much does Castform Qwen3.5-4B inference cost?
Castform lists $0.03 per 1 million input tokens and $0.15 per 1 million output tokens for hosted Qwen3.5-4B inference.
How much does Qwen3.5-4B training cost on Castform?
Castform lists an introductory training rate of $0.10 per 1 million training tokens. Training usage is a separate billing unit from hosted inference.
Does the public Castform endpoint include the Neon retrieval fine-tune?
Not by default. The public rate card identifies the Qwen3.5-4B base-model endpoint. The benchmark article does not publish a stable public model ID for its tuned derivative.
Why is the Castform fine-tune not in AI Pricing Guru Labs?
The exact tuned artifact, full evaluation set, reward weights, stopping rules, and measured usage are not public. Testing the base endpoint would not reproduce the reported specialist result.
Sources and maintenance
Inference and training rates are checked against Castform's pricing page and public pricing API. Benchmark claims are checked against the Neon and Castform case study and public example repository. Sources are watched daily.