TAAFT
Free mode
100% free
Freemium
Free Trial
Deals
Companies
DeepReinforce
Models
Ornith 1.0 35B

Ornith 1.0 35B

By DeepReinforce
Model family: Qwen
Ornith-1.0-35B is the mid-tier model in the Ornith-1.0 family, a 35B MoE architecture post-trained on Qwen 3.5. It is a reasoning model that produces explicit think traces before final answers and emits well-formed function calls via the tool_calls field. The core innovation is a self-improving RL training framework: rather than relying on fixed human-designed harnesses, the model co-learns solution rollouts and the task-specific scaffolds that guide them. Reward is propagated to both stages, enabling per-task orchestration strategies to emerge automatically. The context window is 262,144 tokens. On agentic coding benchmarks, it scores 75.6 on SWE-Bench Verified, 50.4 on SWE-Bench Pro, 69.3 on SWE-Bench Multilingual, and 64.2 on Terminal-Bench 2.1, outperforming Qwen 3.5-397B on Terminal-Bench 2.1 despite being 10x smaller. Compatible with vLLM, SGLang, and HuggingFace Transformers. MIT licensed.
New Coding Gen 2
Released: June 27, 2026

Overview

Ornith-1.0-35B is a 35B Mixture-of-Experts reasoning model for agentic coding, post-trained on Qwen 3.5 using a self-improving RL framework that jointly learns solution rollouts and the task-specific scaffolds guiding them. Supports native function calling and 262K context. Scores 75.6 on SWE-Bench Verified and 64.2 on Terminal-Bench 2.1. MIT licensed.

‍💻Code generation🤔Logical reasoning🤖Ai agents

About DeepReinforce

DeepReinforce is an AI research startup founded by Dr. Jiwei Li, focused on using reinforcement learning to build agentic AI systems for coding and system optimization. They developed GrandCode (ranked #1 in Codeforces live competitions, beating all human grandmasters), Ornith-1.0 (open-source LLMs for agentic coding, 9B–397B parameters), and IterX (agentic code optimizer surpassing NVIDIA's cuBLAS).

Industry: Artificial Intelligence
Location: US
Website: deep-reinforce.com
View Company Profile

Other models from this family

View all models from this family

Related Models

  • AnnouncementGen 2
    Base 1
    Base 1 is a model trained by Base44 specifically for building web applications inside the Base44 platform. Trained on millions of real building sessions, it is designed not only to write code but to make product decisions, anticipate user needs, and act as a technical partner throughout the app creation process.
    ‍💻Code generation💻Web development assistance🛠️App building
    NewCoding
    Released 1d ago
  • Gen 2
    Composer 2
    Composer 2 is Cursor’s frontier coding model optimized for high intelligence per dollar, built to solve long-horizon software engineering tasks with many tool actions.
    💻Coding🔧Code refactoring💻Vibe coding
    Coding
    Released 3mo ago
  • Gen 2
    Voyage code 3
    Text embedding model optimized for code retrieval. Supports Matryoshka embeddings at 256, 512, 1024, and 2048 dimensions with quantized formats (float, int8, uint8, binary, ubinary) for up to 32x storage reduction. Features a 32K-token context window. Outperforms OpenAI text-embedding-3-large by 13.80% on 32 code retrieval datasets covering text-to-code, code-to-code, and docstring-to-code tasks.
    🔎Semantic search🔍Code search📦Code retrieval
    Coding
    Released 1y ago
Last updated: June 30, 2026
0 AIs selected
Clear selection
#
Name
Task