03 Categories & Licenses17 VERIFIED ENGINES

Best AI Coding Models & Agentic IDEs (2026 Benchmark & VRAM Index)

Compare repository-scale AI coding IDEs (Cursor, GitHub Copilot) against self-hosted open-weight coding models (Qwen2.5-Coder-32B, Qwen 2.5 14B, Gemma 2 27B).

CATEGORYCoding & Repo IDEs

BEST LOCAL CODERQwen2.5-Coder-32B

BEST CLOUD IDECursor ($20/mo)

SIBLING HUB URLS:/hubs/categories-licenses /category/coding-ides /category/image-design /category/video-cinema /category/voice-audio /category/research /category/agents /category/writing-copy /category/productivity-tools /category/3d-spatial /license/apache-2

Verified Engines for /category/coding-ides

17 RESULTSClick any model card to open its dedicated /ai/[slug] page.

●CLOUD • 93/100

Claude

Anthropic

Anthropic's assistant, known for long-document reasoning, nuanced writing, and coding, with context windows up to 1M tokens on higher tiers.

BASE ARCHITECTURE / Claude Sonnet / Opus 4.x

SInstant Cloud Web Studio (0 GB VRAM • Free / $20 mo+)

ℹTHE BRUTAL TRUTH

The only consumer assistant with a real 1M context that works. If you read 200+ page documents, this is the one. Otherwise, ChatGPT's ecosystem wins.

API

ILLUSTRATIVE PROMPT

Summarise a 400-page technical document with section-level citations.

PREVIEW NOTE

Illustrative long-context task; not executed here.

1,178

Launch↗

●LOCAL • 94/100

Qwen2.5-Coder-32B-Instruct

Alibaba / Qwen

Alibaba Qwen's flagship open-weight coding model with 32.5B dense parameters, supporting a 128K context window with verified state-of-the-art Python, TypeScript, and multi-language code generation.

BASE ARCHITECTURE / Qwen2.5-Coder-32B-Instruct

BRuns well (61/100) • Q4_K_M (~35 tok/s • 4GB free)

ℹTHE BRUTAL TRUTH

The premier open coding model for local developers. If you have a 24GB GPU, it delivers frontier coding assistance with zero API fees.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Refactor this service into typed modules and generate integration tests.

PREVIEW NOTE

Production-ready typed code implementation with complete unit tests.

1,120

Hugging Face↗

●CLOUD • 90/100

GitHub Copilot

GitHub

The most widely deployed AI coding assistant, embedded across VS Code, JetBrains, and GitHub itself with multi-model routing.

BASE ARCHITECTURE / Multi-model routing

SInstant Cloud Web Studio (0 GB VRAM • Free / $10 mo)

ℹTHE BRUTAL TRUTH

Cheaper than Cursor, but the agentic workflow is still behind. If you're on GitHub Enterprise already it's a no-brainer — otherwise, look at Cursor first.

API

ILLUSTRATIVE PROMPT

Review this pull request for null-safety issues.

PREVIEW NOTE

Illustrative review task; not executed here.

1,102

Launch↗

●CLOUD • 94/100

Cursor

Anysphere

An AI-first code editor with repository context, agent workflows, and selectable model providers. The dominant AI-native IDE by revenue.

BASE ARCHITECTURE / Multiple selectable models

SInstant Cloud Web Studio (0 GB VRAM • Free tier / $20 mo)

ℹTHE BRUTAL TRUTH

It's a wrapper — you're paying $20/mo for orchestration on top of GPT/Claude. Only worth it if it saves you 1+ hour per week. Otherwise, self-host Qwen2.5-Coder-32B.

ILLUSTRATIVE PROMPT

Refactor this module into smaller typed functions and add tests.

PREVIEW NOTE

A proposed refactor plan with file-level edits and test suggestions.

942

Launch↗

●LOCAL • 90/100

Qwen2.5-14B-Instruct

Alibaba / Qwen

Alibaba Qwen's high-efficiency 14.7B dense instruction-tuned model with 128K context, hitting the sweet spot for 12GB–16GB consumer GPUs.

BASE ARCHITECTURE / Qwen2.5-14B-Instruct

SRuns well (85/100) • Q4_K_M (~71 tok/s • 14GB free)

ℹTHE BRUTAL TRUTH

The best quality-per-GB model on 12GB–16GB consumer hardware. Outstanding balance of speed and coding capability under Apache 2.0.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Build a typed API client with retries and unit tests.

PREVIEW NOTE

Complete TypeScript client with retry logic and unit tests.

940

Hugging Face↗

●CLOUD • 94/100

Lovable.dev

Lovable

No-code full-stack AI web app builder. Generates working Supabase backends, React frontends, auth flows, and GitHub PRs directly from chat conversations.

BASE ARCHITECTURE / Lovable Full-Stack Builder (Claude 3.7 + GPT-4o)

SInstant Cloud Web Studio (0 GB VRAM • Free trial · $20/mo)

ℹTHE BRUTAL TRUTH

The most capable full-stack builder available for shipping production-ready web apps without boilerplate. You own the exported GitHub code.

No cardAPI

ILLUSTRATIVE PROMPT

Build an Airbnb clone for pet sitters with booking calendar and Stripe checkout.

PREVIEW NOTE

Working React application connected to Supabase with real-time bookings table.

930

Launch↗

●CLOUD • 95/100

v0 by Vercel

Vercel

Vercel's generative UI builder. Turns natural language prompts into accessible React components, Tailwind CSS styling, and shadcn/ui layouts with 1-click deploy.

BASE ARCHITECTURE / v0 Generative UI Engine

SInstant Cloud Web Studio (0 GB VRAM • Free tier · $20/mo)

ℹTHE BRUTAL TRUTH

The absolute gold standard for scaffolding React and Tailwind UI components. Eliminates hours of CSS tweaking.

No cardAPI

ILLUSTRATIVE PROMPT

Create a modern dark-mode analytics dashboard with revenue charts and user breakdown.

PREVIEW NOTE

Full React component using Tailwind CSS, Lucide icons, and Recharts.

915

Launch↗

●LOCAL • 91/100

Gemma 2 27B

Google DeepMind

Google DeepMind's flagship 27.2B open model featuring alternating local/global attention and post-training alignment, punching above its weight class on single-GPU hardware.

BASE ARCHITECTURE / Google Gemma 2 27B (Instruct)

BRuns well (60/100) • Q4_K_M (~39 tok/s • 6GB free)

ℹTHE BRUTAL TRUTH

Google's 27B architecture punches way above its parameter weight. If 8K context is sufficient for your workload, it is an outstanding 24GB driver.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Implement a rate-limited API endpoint with structured errors.

PREVIEW NOTE

Robust implementation with structured error handling.

890

Hugging Face↗

●CLOUD • 93/100

Windsurf IDE

Codeium

Agentic AI code editor with Cascade flow. Deeply indexes codebases, navigates multi-file refactors, and synchronizes linter errors in real time.

BASE ARCHITECTURE / Cascade Agentic Engine (Claude 3.7 + GPT-4o)

SInstant Cloud Web Studio (0 GB VRAM • Free tier · $15/mo)

ℹTHE BRUTAL TRUTH

The most serious direct contender to Cursor. Cascade's deep integration across terminal output and file diffs is genuinely impressive at $15/mo.

No cardAPI

ILLUSTRATIVE PROMPT

Audit this repository for SQL injection vulnerabilities and refactor database queries to use prepared statements.

PREVIEW NOTE

Systematic multi-file diff applying parametrized queries across 8 repository files.

890

Launch↗

●LOCAL • 89/100

Qwen2.5-Coder-7B-Instruct

Alibaba / Qwen

State-of-the-art 7B code LLM rivaling older 34B models. Fits on virtually any 6GB–8GB GPU with 128K context and Apache 2.0 licensing.

BASE ARCHITECTURE / Qwen2.5-Coder-7B-Instruct

SRuns well (98/100) • Q4_K_M (~141 tok/s • 19GB free)

ℹTHE BRUTAL TRUTH

The absolute king of 8GB GPUs for coding. It beats models four times its size, runs at 80 tok/s on an RTX 3060, and carries Apache 2.0.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Write a SQL migration script to add soft deletes and index created_at on user tables.

PREVIEW NOTE

A PostgreSQL migration script adding deleted_at TIMESTAMP WITH TIME ZONE and a partial index.

840

Hugging Face↗

●LOCAL • 91/100

DeepSeek-Coder-V2-Lite-Instruct

DeepSeek

A lightweight Mixture-of-Experts coding model with 16B total parameters and 2.4B active. Supports 338 programming languages and a 128K context window.

BASE ARCHITECTURE / DeepSeek-Coder-V2-Lite-Instruct (16B MoE)

SRuns well (85/100) • Q4_K_M (~64 tok/s • 13GB free)

ℹTHE BRUTAL TRUTH

The best sub-16GB open coding model for multi-language repos. Blazing fast at 60 tok/s on an RTX 4070.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Build a concurrent WebSocket proxy in Rust with channel reconnection logic and integration tests.

PREVIEW NOTE

A structured Rust module implementing Tokio broadcast channels and graceful client disconnection handling.

765

Hugging Face↗

●LOCAL • 90/100

Codestral-22B

Mistral AI

Mistral's dedicated 22B open-weight code generation and fill-in-the-middle model with 32K context, fluent in over 80 programming languages.

BASE ARCHITECTURE / Codestral-22B-v0.1

ARuns well (77/100) • Q4_K_M (~50 tok/s • 10GB free)

ℹTHE BRUTAL TRUTH

Superb code generation and fill-in-the-middle accuracy, but the MNPL license prevents production commercial SaaS deployment without contacting Mistral.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Implement an LRU cache with O(1) get and put in C++ using std::list and std::unordered_map.

PREVIEW NOTE

An optimized C++17 LRU cache class with lock guards and iterator eviction.

730

Hugging Face↗

●LOCAL • 86/100

StarCoder2-15B

BigCode / ServiceNow

Trained on 4+ trillion tokens across 600+ programming languages from The Stack v2. Transparent, permissively trained code generation workhorse.

BASE ARCHITECTURE / StarCoder2-15B (dense)

SRuns well (85/100) • Q4_K_M (~71 tok/s • 14GB free)

ℹTHE BRUTAL TRUTH

Enterprise legal teams love the clean training dataset provenance. Good autocomplete, but shorter context than modern models.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Write a Python data validator using Pydantic v2 with custom field serializers.

PREVIEW NOTE

A typed Pydantic BaseModel with @field_validator and model_serializer functions.

612

Hugging Face↗

●LOCAL • 82/100

Code Llama 34B Instruct

Meta

Meta's 34B parameter code-specialized model with infilling capabilities and large context. Battle-tested base for enterprise fine-tuning.

BASE ARCHITECTURE / Code Llama 34B Instruct

BRuns well (61/100) • Q4_K_M (~35 tok/s • 4GB free)

ℹTHE BRUTAL TRUTH

Historic milestone, but newer models like Qwen 3.6 27B and Qwen3-Coder beat it soundly while using fewer resources. Only use if tied to existing fine-tunes.

No cardLocalAPIOpen weightsHF

ILLUSTRATIVE PROMPT

Create a Bash script with getopt that parses flags, logs errors to stderr, and verifies disk space.

PREVIEW NOTE

A robust POSIX-compliant shell script with trap cleanup handlers and error codes.

590

Hugging Face↗

●LOCAL • 96/100

Clover 35B A3B (Diffusion LLM)

DiffuSpace / HKU NLP

DiffuSpace's breakthrough Diffusion Language Model (dLLM). Replaces traditional autoregressive token generation with parallel block diffusion denoising across 35B MoE parameters (3B active), shattering SWE-bench Verified coding tasks.

BASE ARCHITECTURE / Clover-35B-A3B-Preview (Block Diffusion MoE)

ARuns well (79/100) • Q4_K_M (~47 tok/s • 9GB free)

ℹTHE BRUTAL TRUTH

The first true demonstration that diffusion language models can beat autoregressive models on real SWE-bench coding. Parallel denoising means complete code functions appear in bursts rather than token-by-token trickles.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Synthesize an entire concurrent network driver in Rust with memory-safe buffer management.

PREVIEW NOTE

Parallel-generated Rust implementation verified through multi-token diffusion steps.

540

Hugging Face↗

●LOCAL • 98/100

Spark X2.6 1.7B

SparkLLM Team / XHToken

XHToken's state-of-the-art 1.7B on-device reasoning and coding model. Features a 1M token context length, configurable thinking effort modes (none/low/high), and outsized performance on SWE-bench and autonomous planning.

BASE ARCHITECTURE / Spark-X2.6-1.7B (Sliding-Attention Transformer)

SRuns well (94/100) • Q4_K_M (~220 tok/s • 22GB free)

ℹTHE BRUTAL TRUTH

The undisputed champion of 1.7B on-device reasoning. Having a controllable thinking budget (none/low/high) plus a 1M token context under Apache 2.0 makes it the best lightweight agent brain available.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Analyze this 20-file repo trace, identify the race condition, and output a minimal patch.

PREVIEW NOTE

Step-by-step thinking trace followed by precise diff patch.

460

Hugging Face↗

●LOCAL • 92/100

Nanbeige 4.3 33B

Nanbeige LLM Lab

Nanbeige's flagship 33B dense language and coding model with a 512K context window. Engineered specifically for agentic terminal workflows, SWE-bench engineering, and professional enterprise co-work under Apache 2.0.

BASE ARCHITECTURE / Nanbeige4.3-33B (Dense Teacher Architecture)

DDecent (36/100) • Q4_K_M (~21 tok/s • 3GB free)

ℹTHE BRUTAL TRUTH

A serious alternative to Qwen2.5-Coder-32B built specifically for autonomous agents that run terminal commands and refactor full repos. If you run SWE-bench agents on a 24GB card, this is a top-tier driver.

No cardLocalAPIOpen weightsHFPermissive

ILLUSTRATIVE PROMPT

Diagnose this failing CI pipeline across 4 microservices and execute terminal fixes.

PREVIEW NOTE

Step-by-step reasoning trace with verified terminal commands and patch files.

410

Hugging Face↗

Frequently Asked Questions · /category/coding-ides

What is the best local open-source coding LLM for 24GB VRAM?

Qwen2.5-Coder-32B (18GB at Q4_K_M, 256K context, Apache 2.0) and Qwen 2.5 14B (17GB at Q4_K_M) lead single-GPU coding benchmarks.

Is Cursor worth $20/mo if I have a 24GB GPU?

Cursor's multi-file tab prediction and repo indexing save hours if you bill $50+/hr, while pairing Continue.dev or Cline with local Qwen3-Coder gives you 100% air-gapped privacy for $0.