Skip to content
← Models
Open weightsAgentica / Together AI/10 trustVerified
DeepCoder-14B-Preview logo

DeepCoder-14B-Preview

14B RL-trained code reasoning model at o3-mini level.

Visit official page →

About

DeepCoder-14B-Preview is a fully open 14B coding model from the Agentica team and Together AI, fine-tuned from DeepSeek-R1-Distilled-Qwen-14B with distributed reinforcement learning over 24K verifiable coding problems. It hits 60.6% Pass@1 on LiveCodeBench, matching o3-mini, and the team open-sourced the dataset, training scripts, and logs alongside the weights.

It is a preview research artifact rather than a polished product — no tool calling, English-only, and the headline scores are team-reported. The MIT license and full training recipe make it one of the most reproducible RL-for-code results published.

Specs

Context window

64K

Pricing

Free quota

Open weights free to download and self-host

API access

Weights

Open weights

Released

April 8, 2025

Benchmarks

  • LiveCodeBench: 60.6% Pass@1 (team-reported)
  • Codeforces rating: 1936, 95.3rd percentile (team-reported)

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

coderesearch

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: Agentica x Together AI, Apr 2025; MIT license with dataset, training scripts, and logs fully open; Preview research artifact rather than a product — no tool calling, English-only; LiveCodeBench 60.6% is team-reported but the reproducible recipe makes the claim auditable; High openness: one of the most reproducible RL-for-code results published. License: MIT — genuinely open.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
8.0/10
Maintenance · recency, releases, issue load
7.0/10
Openness · license + docs
9.0/10
Manual review · sanity read + red-flag check
8.0/10
  • · Agentica x Together AI, Apr 2025; MIT license with dataset, training scripts, and logs fully open
  • · Preview research artifact rather than a product — no tool calling, English-only
  • · LiveCodeBench 60.6% is team-reported but the reproducible recipe makes the claim auditable
  • · High openness: one of the most reproducible RL-for-code results published

Sources