P
Particle-1
Particle-1 is a token-efficient agentic coding model from Particle AI, post-trained on DeepSeek V4.1 Flash.
Specifications
| Lab | Particle Ai |
|---|---|
| Context window | 1M tokens |
| Input price | $0.05/1M |
| Output price | $0.50/1M |
| Release | Oct 2026 |
Strengths
- ~45% fewer output tokens than DeepSeek V4.1 Flash at parity accuracy on Terminal-Bench 2.1, ReactBench, ADE Bench, and DeepSWE 1.1 (vendor evals)
- Independent Plotly analytics exam: 40/43, matching GPT-6.1 Terra at roughly a quarter of the cost
- Ultra-low pricing ($0.05/M input, $0.50/M output) with a 1M-token context window
- General benchmarks hold near base: GPQA Diamond 90.4%, MMLU 86.1%
Weaknesses
- A rank-64 LoRA fine-tune (~1.25% of parameters) of DeepSeek V4.1 Flash, not a from-scratch frontier model
- Weights closed; API-only via Particle or OpenRouter
- Efficiency gains skew to coding and terminal workloads and did not carry over to one independent analysis exam
Best for
- Agentic software engineering
- High-volume coding automation on tight token budgets
- Data engineering pipelines
In Depth: Particle-1
Summary
Particle-1 is an AI model from Particle Ai.
Released in Oct 2026. It takes text and image input and produces text output, with a context window of 1M tokens. Input pricing is $0.05 per 1M tokens and output is $0.50 per 1M tokens on OpenRouter.