OpenRouter model

OpenAI: gpt-oss-20b

gpt-oss-20b is an open-weight 21B parameter model released by OpenAI under the Apache 2.0 license. It uses a Mixture-of-Experts (MoE) architecture with 3.6B active parameters per forward pass, optimized for...

Provider
OpenRouter
Availability
Available
Model ID
openai/gpt-oss-20b
Snapshot
2026-10-07
Evaluation Verdict

OpenAI gpt-oss-20b is a lightweight open-weights coding model (41.0% SWE-bench) designed for single-GPU local workstations and fast edge subagent execution.

OpenAI: gpt-oss-20b Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified
SWE-bench Verified
41%
StandardOpenAI Open Weights Report ↗
LiveCodeBench
38%
StandardLiveCodeBench ↗
LMSYS Coding Elo
1285Elo
Coding ArenaLMSYS Arena ↗
AIME 2024 / 2025
48%
StandardOpenAI Open Weights Report ↗
Pricing ($ / 1M tokens)
$0.1/ $0.3
$0.04 cached readStandard API Pricing

What This Means in Real Software Development

  • Fits on single consumer GPUs (24GB VRAM) via 4-bit/8-bit quantization.
  • Scores 41.0% on SWE-bench Verified without needing cloud API keys.
  • Fast execution for rapid local symbol search and linting.
  • Complete offline air-gapped development support.

Swarm Multi-Agent Topology

Tier 3: Local Subagent WorkerRole: Finder Subagent

Best Paired With: gpt-oss-120b or Claude 3.7 Sonnet

Fast local AST symbol indexing and lint check loops with zero API cost.

Git Worktree Isolation Advantage:

Dispatched across local Git worktrees via Ollama or vLLM.

What Developers & Practitioners Are Saying

Real-world field observations, production strengths, and developer consensus

Field Consensus

"Lightweight open-weights model designed for local workstations and developer laptops."

Verified Real-World Strengths

  • Can be quantized to 4-bit and run locally on Apple Silicon or consumer GPUs
  • Fast local inference without cloud network latency
  • Zero data privacy risk

Production Caveats & Gaps

  • Lower SWE-bench score (28.5%) compared to frontier cloud models
Benchmark Reality Check

What the benchmark tests: Measures localized code generation and basic instruction following.

Production caveat: Best suited for offline autocomplete and local syntax checking.

Primary Sources & Methodology Citations

OpenAI: gpt-oss-20b model overview

OpenAI: gpt-oss-20b is listed in the public Swarm catalog through OpenRouter. This page summarizes the verified catalog fields used by Swarm without claiming real-time access for every provider account.

Provider
OpenRouter
Family
OpenRouter
Context window
131.1K tokens
Thinking support
low · medium · high
Availability status
Available
Catalog ID
openrouter/openai/gpt-oss-20b
  • Text input
  • Text output
  • Reasoning
  • Tools
  • Structured output
  • JSON

Run OpenAI: gpt-oss-20b in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →
bash — swarm run
$swarm ctl model set --provider openrouter --model openrouter/openai/gpt-oss-20b

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign OpenAI: gpt-oss-20b to specialized system roles in ~/.swarm/config.json:

  • Primary / Planner: Orchestrates full task DAGs and multi-stage pipelines.
  • Coder Subagent: Generates code and unit tests inside private Git worktrees.
  • Finder Subagent: Performs read-only code search, symbol discovery, and mapping.
← All OpenRouter models

Public catalog snapshot. Catalog availability is not a real-time provider status or an access guarantee.