Particle AI
Particle-1
Particle AI post-trained DeepSeek V4.1 Flash to solve software engineering tasks with less generated text.
Run this model
particle/particle-1
Use an endpoint supported by this model.
Get a key for this setup →const response = await fetch('https://api.minirouter.sh/v1/chat/completions', { method: 'POST', headers: { Authorization: `Bearer ${process.env.MINIROUTER_KEY}`, 'Content-Type': 'application/json', }, body: JSON.stringify({ model: 'particle/particle-1', messages: [{ role: 'user', content: 'Why is the sky blue?' }], max_tokens: 1024, }),}) const data = await response.json()console.log(data.choices[0].message.content)This route accepts text input. Particle also documents native image support, which is not enabled on this MiniRouter route.
Provider peak-hour pricingOff-peak prices shown include MiniRouter's markup. The provider charges 2× during 08:00–00:30 UTC the following day, every day. Final charges reflect provider-reported usage, including reasoning tokens. Source ↗
Model overview
- Input / 1M
- $0.13125
- All-in MiniRouter rate
- Output / 1M
- $0.525
- All-in MiniRouter rate
- Context
- 1M
- 1,048,576 tokens
- Max output
- 131K
- tokens
- Base model
- DeepSeek V4.1 Flash
- Post-trained by Particle AI
- Training focus
- Software engineering
- Coding and data engineering
- Announced
- 28 Oct 2026
- publisher release
Read the token count beside the pass rate.
Particle's DeepSWE 1.1 evaluation reports fewer generated tokens with a similar task pass rate to its base model. These results describe the publisher's agent harness.
DeepSWE 1.1
Task pass rate
Compare completed work as well as token use.
Publisher-reported accuracy
0 to 100%
- Particle-172.6%
- DeepSeek V4.1 Flash71.2%
View chart data
| Model | Pass rate |
|---|---|
| Particle-1 | 72.57% |
| DeepSeek V4.1 Flash | 71.15% |
Mean output tokens per task
- Particle-1
- 59,580
- DeepSeek V4.1 Flash
- 83,246
View chart data
| Model | Mean output tokens |
|---|---|
| Particle-1 | 59,580 |
| DeepSeek V4.1 Flash | 83,246 |
Keep the same task and change the model.
- [ 01 ]
Hold the harness steady
Keep prompts, tools and task budgets the same when comparing with another model.
- [ 02 ]
Count successful outcomes
Track completed tasks, retries and total spend. A shorter answer alone does not establish better value.
- [ 03 ]
Inspect what was skipped
Check that concise runs retain the tests, explanations and edge cases your workflow needs.
Use the exact model ID and current route limits.
- [ 01 ]
Pin particle/particle-1
Use the full MiniRouter model ID in your request and an explicit output-token cap.
- [ 02 ]
Validate structured data
Particle does not advertise schema-guaranteed structured output. Validate any JSON before using it.
- [ 03 ]
Budget for reasoning
Reasoning contributes to generated-token usage. Include it when comparing the cost of a completed task.
One route, standard clients.
https://api.minirouter.sh/v1particle/particle-1/v1/chat/completions/v1/messages/v1/responsesConfigured providers. Priority follows the live catalog order.
| Provider | Priority | Input / 1M | Output / 1M | Parameters | Access |
|---|---|---|---|---|---|
vercel | 1 | $0.13125 | $0.525 | max_tokens, temperature, stop, tools, tool_choice, reasoning, include_reasoning | MiniRouter balance |