Choosing a SynapseX model in Chat
SynapseX Chat gives you six first-party LLMs and lets you change the model per conversation. This page tells you which one to pick for the work in front of you, and what that choice costs you against your daily message counters.
How do I switch models?
Click the model chip in the composer, or press Cmd+Shift+P to open the picker. The choice is saved on the conversation, so an old chat reopens on the model it was using. To keep the ones you use most at the top of the list, mark them as favourites in Settings > Models.
The default for a new chat is SynapseX Atlas (synapsex:synapsex-atlas-7b-v1).
Which model should I use?
| Model | Id | Built for | Context | Max output | Input / 1M | Output / 1M |
|---|---|---|---|---|---|---|
| Atlas | synapsex:synapsex-atlas-7b-v1 | General tasks and high volume; compact and fast. This is the default. | 128,000 | 16,000 | $0.10 | $0.40 |
| Forge | synapsex:synapsex-forge-code-14b-v1 | Code generation and review. | 256,000 | 8,000 | $0.40 | $1.20 |
| Vault | synapsex:synapsex-vault-enterprise-v1 | Enterprise long context and compliance. | 131,000 | 16,000 | $0.35 | $1.40 |
| Nexus | synapsex:synapsex-nexus-v2 | Fast multi-agent orchestration across code, chat and tools. | 131,000 | 8,000 | $0.10 | $0.40 |
| Oracle | synapsex:synapsex-oracle-v1 | Advanced reasoning on complex tasks. | 131,000 | 8,000 | $0.75 | $3.00 |
| Quasar | synapsex:synapsex-quasar-v1 | Quantum computing research and frontier reasoning. | 200,000 | 32,000 | $3.50 | $14.00 |
Context and max output are in tokens; prices are US dollars per one million tokens.
This is the SynapseX Chat catalog and its prices. It is not the Platform API price list. The API serves a different, longer catalog with different ids, different context windows and its own rates — see Models and pricing. The two products also bill in different credit units, so never convert a number from one page into the other: Two products, two accounts.
Pick this when
- Everyday questions, lots of them — Atlas. It is the default for a reason: fastest and cheapest per token.
- Reviewing or writing code — Forge. It has the largest context window of the six (256,000 tokens), so a big file plus its tests still fits.
- Long documents, contracts, compliance material — Vault.
- Chats where you expect the assistant to call tools — Nexus.
- A hard problem where the first answer usually is not good enough — Oracle. Consider pairing it with Turbo.
- Quantum and frontier research work — Quasar. It has the largest output budget (32,000 tokens), which matters for long derivations.
What does the choice cost me?
Two daily counters apply, both resetting at UTC midnight:
- 1,000 messages per day for a signed-in account, across all models.
- 500 messages per day on an extra counter that only models outside the free set consume. Atlas and Forge are in the free set; Vault, Nexus, Oracle and Quasar consume this second counter as well as the first.
| Model | Consumes the 500/day pro-model counter? |
|---|---|
| Atlas | No |
| Forge | No |
| Vault | Yes |
| Nexus | Yes |
| Oracle | Yes |
| Quasar | Yes |
So a day of heavy Oracle use can hit the 500 ceiling long before the 1,000 one. If you are working through a long session and do not need frontier reasoning, staying on Atlas or Forge keeps the pro-model counter untouched. Full details, plus your monthly plan quota and every error code, are in Chat limits and errors.
Verify the list yourself
The catalogue is served publicly, so you can always check what is selectable right now:
curl -s https://chat.synapsex.ai/api/modelsEach entry in the returned models array carries the fields you need (shown
abridged — the real entries carry more fields than these):
{
"id": "synapsex:synapsex-atlas-7b-v1",
"name": "SynapseX Atlas",
"description": "Compact, fast model for general tasks and high volume.",
"contextWindow": 128000,
"inputCost": 0.1,
"outputCost": 0.4,
"priceUnit": "per 1M tokens"
}The id is the value the app stores on the conversation. The synapsex: prefix is the provider; the part after it is the deployment name.
The same model is spelled differently on each surface and the spellings are not interchangeable — the full table is on Two products, two accounts.
The picker is not gated by your plan. Every signed-in account sees all six models and can switch a conversation to any of them. What your plan changes is your monthly quota, not which entry you may select.
Related
- Turbo — spend more compute per turn instead of switching to a bigger model.
- Chat limits and errors — the counters and the 402 / 429 responses.
- Chat plans — the monthly message and token allowances.