Particle-1

Multimodal
Particle AI

Particle-1 is Particle AI's first model, a reinforcement-learning post-train of DeepSeek V4.1 Flash (rank-64 LoRA over the experts in all layers, trained with GSPO in Cascade RL stages on internal coding and data-engineering environments). It targets agentic software engineering and is tuned to solve tasks with fewer tokens: Particle reports 28.4% to 52.2% fewer output tokens and 34.4% to 51.6% lower cost per task than the base model on Terminal-Bench 2.1, ReactBench, ADE Bench, and DeepSWE 1.1, with pass@1 within the base model's confidence intervals (72.57% aggregate task pass rate vs 71.15% for DeepSeek V4.1 Flash). Text and image input, text output, reasoning with effort control, OpenAI-compatible tool calling, 1,048,576-token context and 131,072 max output tokens. Weights are not released; it is available through the Particle API and OpenRouter (particle-ai/particle-1) at $0.25 input, $0.03 cached input, and $1.00 output per million tokens (peak; off-peak 50% off, 20:30 to 04:00 America/New_York).

Key Specifications

Parameters
-
Context
1.0M
Release Date
October 28, 2026
Average Score
88.6%

Timeline

Key dates in the model's history
Announcement
October 28, 2026
Last Update
October 31, 2026
Today
November 20, 2026

Technical Specifications

Parameters
-
Training Tokens
-
Knowledge Cutoff
-
Family
-
Fine-tuned from
deepseek-v4-1-flash
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$0.25
Output (per 1M tokens)
$1.00
Max Input Tokens
1.0M
Max Output Tokens
131.1K
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

General Knowledge

Tests on general knowledge and understanding
MMLU
Standard MMLU. Self-reported in the Particle-1 launch blog (base DeepSeek V4.1 Flash: 85.98%). • Self-reported
86.1%

Reasoning

Logical reasoning and analysis
GPQA
GPQA Diamond. Self-reported in the Particle-1 launch blog (base DeepSeek V4.1 Flash: 90.91%). • Self-reported
90.4%

Other Tests

Specialized benchmarks
BabyVision
With tools. Self-reported in the Particle-1 launch blog (base DeepSeek V4.1 Flash: 89.60%). • Self-reported
89.2%

License & Metadata

License
proprietary
Announcement Date
October 28, 2026
Last Updated
October 31, 2026

Compare Particle-1

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.