Tasks and inputs
Particle-1 is listed for Text generation, Coding & Code Generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
Particle-1 is Particle's first software-engineering model, an RL-tuned DeepSeek V4.1 Flash that aims to solve coding tasks with fewer tokens.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: Particle lists $0.30/1M uncached input, $0.006/1M cached, $1.20/1M output; some listings show $0.25/$1.00. Confirm on your account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Can a LoRA adapter make a coding model cheaper without making it worse? Particle-1 is Particle's answer. It is the company's first software-engineering model, released on 28 October 2026, and it is not a new base model at all. It is DeepSeek V4.1 Flash with a reinforcement-learning layer on top.
The training details are unusually specific. Particle used GSPO, a policy-optimisation method, through a rank-64 LoRA adapter that touches about 1.25% of the parameters. The objective was efficiency: solve the same tasks, spend fewer tokens.
On average, Particle reports about 45% fewer tokens than the base model. Across four software-engineering benchmarks the published range is 28 to 52% fewer output tokens and 34 to 52% lower cost per task, while pass@1 stays within the base model's confidence intervals. In plain terms: roughly the same success rate, noticeably smaller bills. These are vendor-reported figures.
| Item | Detail |
|---|---|
| API ID | particle-1 |
| Released | 28 October 2026 |
| Base | DeepSeek V4.1 Flash + rank-64 LoRA |
| Context | 1,048,576 tokens (listed by hosts) |
| Max output | 131,072 tokens (listed by hosts) |
| Reasoning / tools | Yes / yes |
| Indicative price | $0.30 input, $0.006 cached, $1.20 output per 1M |
First, price. Particle prices it at $0.30 uncached input, $0.006 cached and $1.20 output, matching the base model. Some third-party listings show $0.25 and $1.00 instead. Check your Particle or OpenRouter account.
Second, context. The million-token window and 131K output cap come from host listings, while Particle's own benchmark runs used a 512K limit. If you rely on very long repositories, test at the length you need.
Third, images. Hosts list image input; Particle does not mention it. Assume text-only until you verify otherwise.
Particle-1 is available through Particle's API and OpenRouter. The pitch is narrow and honest: same capability tier as the base, lower spend on long agent runs. If you already run V4.1 Flash, switching is a small experiment.
Day one, replay twenty resolved tickets from your repository against both models in the same harness. Day two, compare pass rates, total tokens and dollar cost per resolved ticket. If the pass rate holds and cost falls by a quarter or more, switch. If Particle-1 starts failing on refactors that span many files, that is the shorter-reasoning trade biting. Other options live under code generation models.
Compare the model on the work you need to complete.
Particle-1 is listed for Text generation, Coding & Code Generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 1.05M context and 131.07K maximum output. The listed route has starting rates of $0.30 input and $1.20 output per million tokens. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Gemini 4 Hydrogen is Google's fast, lower-cost Gemini 4 model for everyday coding and agentic execution, with a 1M context, xhigh thinking and a free tier.
OpenAI
GPT-6 Cyber is OpenAI's restricted security model for authorised vulnerability investigation, reproduction and fix verification, for Daybreak Red customers.
Microsoft
MAI-Cyber-1.1-Flash is Microsoft AI's vulnerability-discovery model, available only inside its MDASH scanning harness to verified security customers.
recursive
Recursive Flint is an open-weight 39.8B MoE coding-agent model with 2.5B active parameters, free to try via OpenCode and OpenRouter.
Deepreinforce
Ornith-1.5 122B is an open MIT-licensed MoE coding and agent model with a 262K context, reasoning by default and single-80GB-GPU serving.
MiniMax
MiniMax M3.1 Flash is MiniMax's GA open-weight multimodal MoE for coding agents: 428B total, 23B active, 1M context, from $0.15 per 1M input tokens.