What are the verified coding and reasoning benchmarks for Inkling?

Inkling achieves SWE-bench Verified: 54.5% (Standard), LiveCodeBench: 47%. Verified sources: Artificial Analysis Intelligence Index v4.3.2 (Artificial Analysis).

What is the recommended Swarm multi-agent topology for Inkling?

Swarm recommends Inkling as Finder Subagent (Tier 2: Code Search & Discovery Specialist). Rapidly maps file paths and dependency graphs for the lead planner. Best paired with: Claude Opus 5.5 or GPT-6 Astra.

How does Inkling compare to rival frontier coding models on SWE-bench and pricing?

Inkling achieves SWE-bench Verified: 54.5% at $0.15/$0.6 per 1M tokens. Direct peer comparison: NVIDIA Nemotron 3 Ultra (58.2% SWE-bench, $0.5/$1.5); MiniMax-M3 (58% SWE-bench, $0.25/$1); Qwen 3.8 27B (62% SWE-bench, $0.2/$0.8); K2 Horizon 375B A23B (59.8% SWE-bench, $0.4/$1.6).

Does Swarm isolate Inkling code edits from other agents?

Yes. Swarm provisions isolated Git worktrees for every session and subagent, preventing file race conditions and keeping your main working tree clean until verification passes.

Can Inkling be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure Inkling as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

Fireworks AI model

Inkling

Q: How do I run Inkling with Swarm?

Configure Inkling via CLI using: swarm ctl model set --provider fireworks --model fireworks/inkling, or switch models interactively inside the terminal TUI with /models.

Inkling - Thinking Machines multimodal (audio+vision) MoE. Inkling is the first open-weights model released by Thinking Machines Lab. It is a 975B Mixture-of-Experts with 41B active parameters, trained natively across text, image, and audio. Built as a broad generalist foundation for fine-tuning, with controllable thinking effort.

Provider: Fireworks AI
Availability: Available
Model ID: inkling
Snapshot: 2026-10-07

Inkling Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

54.5%

StandardIndependent Evals ↗

LiveCodeBench

47%

ThinkingLiveCodeBench ↗

LMSYS Coding Elo

1305Elo

Coding ArenaLMSYS Arena ↗

Pricing ($ / 1M tokens)

$0.15/ $0.6

$0.03 cached readStandard API Pricing

What This Means in Real Software Development

Achieves 25 on the Artificial Analysis Intelligence Index v4.3.2.
Scores 29 on the Finance & Accounting Index.
Generates 95 output tokens per second at an ultra-low $0.15 cost per task.
Optimized for high-speed grep, semantic AST search, and fast symbol mapping.

Swarm Multi-Agent Topology

Tier 2: Code Search & Discovery SpecialistRole: Finder Subagent

Best Paired With: Claude Opus 5.5 or GPT-6 Astra

Rapidly maps file paths and dependency graphs for the lead planner.

Git Worktree Isolation Advantage:

Operates read-only across whole worktrees with zero risk of dirty state.

What Developers & Practitioners Are Saying

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

Field Consensus

"A fast, lightweight research subagent. Ideal for scanning large repositories to identify relevant symbols before invoking heavier reasoning models."

Verified Real-World Strengths

Fast 95 TPS generation at low cost ($0.15/task)
Efficient grep and file discovery summaries
Clean concise answers without unnecessary prose

Production Caveats & Gaps

Not intended for complex multi-file architectural refactoring
Lower score on deep math competitions

Primary Sources & Methodology Citations

Artificial Analysis Intelligence Index v4.3.2 — Artificial Analysis (September 2026)

Head-to-Head Frontier Comparisons: Inkling vs Rivals

Standardized, side-by-side benchmark & economics comparison against peer models

Flagship Comparative Matrix

Model	Role / Tier	SWE-bench	LiveCodeBench	LMSYS Elo	AIME	Input / Output
InklingFireworks AICurrent	Tier 2: Code Search & Discovery Specialist	54.5%	47%	1305	—	$0.15 / $0.6
NVIDIA Nemotron 3 UltraNVIDIA Compare Inkling vs NVIDIA Nemotron 3 Ultra →	Tier 2: Accelerated Systems Specialist	58.2%	53%	1365	77%	$0.5 / $1.5
MiniMax-M3MiniMax Compare Inkling vs MiniMax-M3 →	Tier 2: Documentation & Test Specialist	58%	50.5%	1320	—	$0.25 / $1
Qwen 3.8 27BQwen Compare Inkling vs Qwen 3.8 27B →	Tier 2: Open-Weights Specialist	62%	54.5%	1342	—	$0.2 / $0.8
K2 Horizon 375B A23BIndependent Compare Inkling vs K2 Horizon 375B A23B →	Tier 2: MoE Systems Specialist	59.8%	52%	1330	—	$0.4 / $1.6

Explore All AI Model Comparisons Hub (Inkling & Flagship Peers) →

Run Inkling in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider fireworks --model fireworks/inkling

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign Inkling 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.

System Agent Roles →Why Swarm Architecture →

← All Fireworks AI models

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