Rnj-1 Instruct

Elite agentic coding model with advanced tool use and STEM reasoning capabilities

About model

Rnj-1 Instruct is Essential AI's elite 8B agentic coding model, delivering world-class software engineering capabilities that rival models 10x its size. With 20.8% on SWE-bench Verified, exceptional tool use (62.2% BFCL), and a 32K context window, it excels at autonomous coding agents, iterative problem-solving, and multi-step technical workflows. Released under Apache 2.0, Rnj-1 Instruct is deliberately designed for community extension—limited post-training preserves flexibility for domain specialization and fine-tuning.

SWE-bench Verified

20.8%

Elite agentic coding performance, outperforming models 10x larger on real-world software tasks.

Efficient Scale

Compact yet powerful architecture optimized for code, STEM reasoning, and tool orchestration.

BFCL Tool Use

62.2%

Superior function calling and API integration for building autonomous agents and automation systems.

Model key capabilities

Elite Agentic Coding: Industry-leading performance on SWE-bench, resolving PRs, optimizing codebases, and writing tests
Advanced Tool Use: Native function calling, Hermes-format parsing, and stable execution across multi-step pipelines
Mathematics & STEM: Strong capabilities across competition math (AIME), scientific reasoning, and structured problem solving
Built for Extension: Limited post-training enables community fine-tuning for specialized domains and tasks

Quickstart guides

RAG

Building a RAG Workflow

Agents

Agent Workflows

Apps

Next.js Chat Quickstart

Performance benchmarks

Model	GPQA Diamond	HLE	LiveCodeBench	MATH500	SWE-bench verified
Rnj-1 Instruct	38.9%			88.4%
Related open-source models
Competitor closed-source models
Claude Opus 4.6	90.5%	34.2%			78.7%
OpenAI o3	83.3%	24.9%		99.2%	62.3%
OpenAI o1	76.8%			96.4%	48.9%
GPT-4o	49.2%	2.7%	32.3%	89.3%	31.0%

API usage

cURL
Python
Typescript

Endpoint:

essentialai/rnj-1-instruct

curl -X POST "https://api.together.xyz/v1/chat/completions" \
  -H "Authorization: Bearer $TOGETHER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "essentialai/rnj-1-instruct",
    "messages": [
      {
        "role": "user",
        "content": "What are some fun things to do in New York?"
      }
    ]
}'

from together import Together

client = Together()

response = client.chat.completions.create(
  model="essentialai/rnj-1-instruct",
  messages=[
    {
      "role": "user",
      "content": "What are some fun things to do in New York?"
    }
  ]
)
print(response.choices[0].message.content)

import Together from 'together-ai';
const together = new Together();

const completion = await together.chat.completions.create({
  model: 'essentialai/rnj-1-instruct',
  messages: [
    {
      role: 'user',
      content: 'What are some fun things to do in New York?'
     }
  ],
});

console.log(completion.choices[0].message.content);

Model card
Architecture Overview:
• 8.3B parameter instruction-tuned model built on Rnj-1 base with 32K context window supporting extended reasoning chains.
• Post-trained with 150B-token supervised fine-tuning stage optimized for instruction following, agentic behavior, and tool use.
• Specialized for multi-turn interactions with strong fill-in-the-middle (FIM) capabilities scoring 86.21% on HE-FIM-Python.

Training Methodology:
• Trained using Muon optimizer throughout all phases with targeted data distributions for reasoning and agentic abilities.
• Post-training inspired by long context mid-training with YaRN, Nemotron approaches, and agentic environment simulations.
• Deliberately limited post-training to preserve flexibility for community specialization and domain adaptation.

Performance Characteristics:
• Elite agentic coding: 20.8% SWE-bench Verified, 49.0% Performance-Enamel, outperforming comparable models by order of magnitude.
• Advanced code generation: 83.5% HumanEval+, 75.7% MBPP+, 57.1% BigCodeBench, 35.2% LiveCodeBench v6.
• Superior tool use: 62.2% BFCL v3, exceptional function calling and API integration capabilities.
• Strong mathematics: 92.6% GSM8K, 90.8% Minerva-Math, 43.3% AIME'25, competitive on olympiad-level problems.
• Robust scientific reasoning: 38.9% GPQA-Diamond, 30.2% SuperGPQA, 76.7% MMLU-STEM across physics, chemistry, biology.
‍
Applications & use cases
Agentic Development:
• Creating AI assistants that iteratively solve software engineering tasks, resolve PRs, and fix security vulnerabilities.
• Developing agents for performance optimization using profilers and iterative code improvement workflows.

Code Generation & Assistance:
• Powering intelligent code completion and generation across multiple programming languages.
• Creating interactive coding assistants for data analysis, visualization, and end-to-end application development.
• Building developer tools with strong tool-calling capabilities for API integrations and system interactions.

Technical Problem Solving:
• Mathematical problem solving systems for education, research, and computational tasks.
• Scientific reasoning applications requiring long-context understanding across STEM domains.
• RAG systems and knowledge bases for technical documentation, troubleshooting, and support workflows.

Enterprise Applications:
• Internal coding assistants integrated with company codebases and workflows.
• Automated code review and quality improvement systems leveraging agentic capabilities.
• Technical support automation combining tool use, code execution, and reasoning for complex queries.
‍

Related models

Model specifications

Model data

Model provider
Essential AI
Type
Chat
Main use cases
Chat
Deployment
Serverless
Monthly Reserved
Endpoint
essentialai/rnj-1-instruct
Parameters
8B
Context length
32K
Input price
$0.15 / 1M tokens
Output price
$0.15 / 1M tokens
Input modalities
Text
Output modalities
Text

Released
December 3, 2025
Quantization level
BF16
External link
Provider docs
Category
Chat

Run in Playground

Quickstart docs

Deploy model

Rnj-1 Instruct

About model

API usage

Model card

Applications & use cases