What are the verified coding and reasoning benchmarks for DeepSeek: R1?

DeepSeek: R1 achieves SWE-bench Verified: 49.2% (Standard (Pass@1)), LiveCodeBench: 49.8%, AIME: 79.8%. Verified sources: DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (DeepSeek AI (arXiv:2501.12948)).

How does DeepSeek: R1 compare to rival frontier coding models on SWE-bench and pricing?

DeepSeek: R1 achieves SWE-bench Verified: 49.2% at $0.55/$2.19 per 1M tokens. Direct peer comparison: GLM 5.3 (56.4% SWE-bench, $0.35/$1.05); OpenAI o4-mini (56.8% SWE-bench, $1.2/$4.8); Mistral Devstral 2512 (59.2% SWE-bench, $0.5/$1.5); DeepSeek V4.1 Flash (56.2% SWE-bench, $0.3/$0.9).

Does Swarm isolate DeepSeek: R1 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 DeepSeek: R1 be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure DeepSeek: R1 as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

OpenRouter model

DeepSeek: R1

Q: What is the recommended Swarm multi-agent topology for DeepSeek: R1?

Swarm recommends DeepSeek: R1 as Coder Subagent (Tier 2: Deep Algorithmic Specialist). Excellent for complex algorithm design and isolated math/logic verification. Use for intricate standalone jobs within a Task Program. Best paired with: Gemini 3.8 Flash for fast search and linting.

Q: How do I run DeepSeek: R1 with Swarm?

Configure DeepSeek: R1 via CLI using: swarm ctl model set --provider openrouter --model openrouter/deepseek/deepseek-r1, or switch models interactively inside the terminal TUI with /models.

DeepSeek R1 is here: Performance on par with [OpenAI o1](/openai/o1), but open-sourced and with fully open reasoning tokens. It's 671B parameters in size, with 37B active in an inference pass....

Provider: OpenRouter
Availability: Available
Model ID: deepseek/deepseek-r1
Snapshot: 2026-10-07

DeepSeek: R1 Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

49.2%

Standard (Pass@1)DeepSeek-R1 Technical Report ↗

LiveCodeBench

49.8%

StandardLiveCodeBench Leaderboard ↗

LMSYS Coding Elo

1360Elo

Coding ArenaLMSYS Arena ↗

AIME 2024 / 2025

79.8%

Pass@1DeepSeek-R1 Technical Report ↗

Pricing ($ / 1M tokens)

$0.55/ $2.19

$0.14 cached readStandard API Pricing

What This Means in Real Software Development

Trained with large-scale reinforcement learning, producing comprehensive internal thinking trajectories visible to the operator.
Achieves 49.2% on SWE-bench Verified and 97.3% on MATH-500, matching proprietary frontier reasoning models.
Extremely economical API pricing ($0.55/M input, $2.19/M output) and fully downloadable for self-hosted private on-premise execution.
Best suited for deep reasoning passes; slower token throughput (~35 TPS) compared to flash models.

Swarm Multi-Agent Topology

Tier 2: Deep Algorithmic SpecialistRole: Coder Subagent

Best Paired With: Gemini 3.8 Flash for fast search and linting

Excellent for complex algorithm design and isolated math/logic verification. Use for intricate standalone jobs within a Task Program.

Git Worktree Isolation Advantage:

Can be self-hosted locally on private GPU clusters, with Swarms headless daemon (swarmd) dispatching local worktrees without any outbound cloud telemetry.

What Developers & Practitioners Are Saying

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

Field Consensus

"The open-weights reasoning sensation that shook the AI industry. Rivals o1 in pure math and code deduction at 1/20th the API price."

Verified Real-World Strengths

Phenomenal 79.8% AIME and 71.5% GPQA Diamond at just $0.55/M input
Transparent `<think>` reasoning tokens allowing developers to inspect logic
Exceptional value for isolated, compute-intensive algorithm subagents

Production Caveats & Gaps

Occasional language switching (spontaneous Chinese/English mix) during thinking
Verbosity can lead to high output token counts if unconstrained
Can struggle with rigid formatting instructions if thinking exhausts output limit

Primary Sources & Methodology Citations

DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning — DeepSeek AI (arXiv:2501.12948) (January 2025)

Head-to-Head Frontier Comparisons: DeepSeek: R1 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
DeepSeek: R1OpenRouterCurrent	Tier 2: Deep Algorithmic Specialist	49.2%	49.8%	1360	79.8%	$0.55 / $2.19
GLM 5.3Zhipu AI Compare DeepSeek: R1 vs GLM 5.3 →	Tier 2: High-Performance Open MoE	56.4%	51%	1350	75%	$0.35 / $1.05
OpenAI o4-miniOpenAI Compare DeepSeek: R1 vs OpenAI o4-mini →	Tier 2: High-Speed Deductive Engine	56.8%	73.4%	1385	90.1%	$1.2 / $4.8
Mistral Devstral 2512Mistral Compare DeepSeek: R1 vs Mistral Devstral 2512 →	Tier 2: Dedicated Autonomous Coder	59.2%	54%	1360	71.5%	$0.5 / $1.5
DeepSeek V4.1 FlashDeepSeek Compare DeepSeek: R1 vs DeepSeek V4.1 Flash →	Tier 3: Refined Open Subagent	56.2%	52%	1360	74.5%	$0.3 / $0.9

Explore All AI Model Comparisons Hub (DeepSeek: R1 & Flagship Peers) →

Run DeepSeek: R1 in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider openrouter --model openrouter/deepseek/deepseek-r1

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign DeepSeek: R1 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 OpenRouter models

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