Curated list of free LLM APIs, coding copilots, AI IDEs, agents, and infrastructure tools for building real AI applications.

• ✅ Free GPT-6 / Claude 5.5 / Gemini 3.5 API access
• 🤖 Coding copilots and AI-native IDEs (Cursor, Trae, Windsurf)
• 💰 Cheapest AI APIs ($0.08-0.50 per 1M tokens)
• 📚 RAG stack tools (vector DBs, embeddings, frameworks)
• 🎯 Agent frameworks and automation tools
• 🔒 Local models for privacy (Ollama, Llama, Qwen)
• 🏗️ Production-ready stack configurations
• 🆕 Claude Sonnet 5.5, Claude Opus 5.5, Haiku 5.5 — GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna — Gemini 3.5 Flash, Gemini 4 Argon — GitHub Copilot AI Credits — Windsurf Max — Trae Ultra — OpenCode 167k⭐ — Xiaomi MiMo V2.5 Pro

Goal: Help developers build AI apps without paying $200/month.

Most AI tool lists are:

• ❌ Outdated (prices/limits from 2023)
• ❌ Filled with affiliate links and sponsored placements
• ❌ General-purpose directories with no developer focus
• ❌ Missing production-critical details (rate limits, commercial use, architecture patterns)

This repo focuses only on:

• ✅ Tools developers actually use in production
• ✅ Generous free tiers (no "5 requests then paywall")
• ✅ Production-capable models (SWE-bench verified, not toys)
• ✅ Real infrastructure (APIs, hosting, vector DBs, not just chatbots)
• ✅ Minimal fluff, maximum utility

Unlike: awesome-ai (general list), ai-collection (marketing focus), toolify (affiliate-heavy)

This is for: Builders who want to ship AI features this week.

If this repo helped you build something or saved you money:

⭐ Star this repo — it helps more builders discover free AI resources.

[🔄 Share with your team] — spread the knowledge.

📝 Contribute — found a new free tier? Updated pricing? PRs welcome!

2026-10-05 (5/10/26)

• 🚀 October 2026 Model Generation Leap: Updated baseline frontier models to the latest production architectures across all tables, IDE profiles, CLI tools, and API pricing matrices:
  • OpenAI: GPT-6 Astra (flagship reasoning), GPT-6.1 Sol ($2.00/$10.00 per 1M, high-efficiency production agent), and GPT-6 Luna ($0.30/$1.20 per 1M lightweight tier).
  • Anthropic: Claude Opus 5.5 ($5/$25, deep reasoning & sandbox security), Claude Sonnet 5.5 ($2/$10, 70.6% Terminal-Bench 4.0, 30% fewer tokens/task), and Claude Haiku 5.5.
  • Google: Gemini 3.5 Flash & 3.5 Flash-Lite (sub-100ms high-throughput defaults) and Gemini 4 Argon preview.
  • Replaced legacy 2025/early-2026 SWE-bench benchmarks in comparison notes with Terminal-Bench 4.0, DeepSWE v1.1, and OSWorld 2.0.
• 🧹 Removed Fabricated & Phantom Tool Entries:
  • Completely purged non-existent / fictitious repositories: Antigravity (google-deepmind/antigravity), MemoryPalace (milla-jovovich/mempalace), AWS Kiro (kiro.dev), and QuantFlow Pilot (qf-studio/pilot).
  • Removed mock benchmark submission "Big Pickle" from OpenCode Zen and agent stacks.
  • Restored verified first-party documentation for Amazon Q Developer (50 interactions/month free tier with AWS Builder ID, powered by Bedrock).
• 🛠️ Tool-Specific Architectural Corrections:
  • Trae: Accurately documented ByteDance's native Doubao model family in China alongside official routing to Claude 3.5/3.7 Sonnet and GPT-4o for international developers (removed false claims about DeepSeek replacing Claude).
  • Atlassian Rovo Dev CLI: Removed from free developer stack; clarified that Rovo is an enterprise SaaS add-on requiring an active paid Atlassian Cloud organization.
• 💰 Pricing & Expired Promotion Refreshes:
  • GitHub Copilot: Summer promotion expired Sep 1, 2026; reverted to standard monthly AI Credits (Business: 1,900 credits / $19, Enterprise: 3,900 credits / $39).
  • Cursor: Shifted Bugbot from flat $40/seat to per-run PR review usage pricing (~$1.00–$1.50/run); split Teams into Teams Standard ($40/seat) & Teams Premium ($120/seat with 5x usage pool); clarified Pro+ ($60/mo) includes $70 in third-party model credits.
  • Qoder: Post-promotional standard pricing settled at Pro $30/mo (2,000 credits, or CNY 59/mo on Qoder CN) and Teams $40/seat (3,000 credits).
  • Xiaomi MiMo: Permanent API pricing flattened to $1.00/1M in, $3.00/1M out, and $0.20/1M cache; fixed Max tier credits typo from 82B to 3B Credits (2.5B–3.5B range).
• 🛑 Service Sunsets & License Corrections:
  • GitHub Models: Documented official and permanent retirement on July 30, 2026.
  • Cohere: Marked strictly as ❌ Non-commercial (evaluation only) per Trial API license terms.
  • Cerebras: Unified free tier quota to 1.5M tokens/day across all tables and matrices.
  • free-coding-models (FCM): Updated live catalog count to 271 models across 25 providers.
  • Synchronized and validated all website data in website/src/data/tools.ts and website/src/data/stacks.ts.
• 🔗 Anchor Links & Navigation Fixes:
  • Replaced dead in-page # anchor links for PR review and CLI tools (Bito, Sourcery, ForgeCode, Goose, OhMyPi) with direct links to official repositories and websites.
  • Corrected invalid URLs for OpenAI image models (pointing to verified DALL-E 3 documentation) and aligned CLI headings to official sites.
• 🧹 Purged Leftover Prompt Artifacts:
  • Removed all internal prompt artifact references ("ExamAi") across RAG architecture diagrams, scaling strategy tables, and stack configs in both README.md and website/src/data/tools.ts.
• ⚖️ Cross-Table Consistency & Math Corrections:
  • Standardized MiniMax model version to MiniMax M2.5 across all API pricing, use-case recommendations, and OpenCode configurations.
  • Aligned Recraft free tier quota to verified 30 credits/day across all tables.
  • Standardized Sora references to Sora 3.
  • Corrected Leonardo.Ai token generation math to reflect realistic output (~8–15 images from 150 daily tokens).
• ⚙️ Provider Categorization & CLI Accuracy:
  • Relocated Vercel AI Gateway ($5/mo credit) from Fully Free Providers to Providers with Trial Credits.
  • Corrected CLI flag typo (--fiable → --reliable) in free-coding-models usage documentation.
  • Removed rogue footer promotional link.

2026-06-25

• 🔄 Major model verification and name alignment: Migrated old placeholders to official Claude Fable 5, Claude Opus 4.8, and GPT-5.5 (Instant/Thinking/Pro) architectures.

2026-06-16

• 🆕 Added OpenCode (167k⭐ OSS CLI), Xiaomi MiMo Token Plan (Chinese coding subscription)
• 🧹 Removed weak/no-longer-free items from Free LLM providers: Cohere (non-commercial only), GitHub Models (Copilot-required; retired July 30, 2026), SambaNova/Hyperbolic (trial-only), HuggingFace (~$0.10/mo), Vercel ($5/mo), Mistral Codestral, Together AI, iFlow (7-day key), Perplexity API
• 🔄 Updated Gemini CLI entry: 3.1 Pro is paid-only; 3 Flash is the free tier (1,500 req/day)
• 🔄 Pricing refresh: Windsurf (Mar 18), Trae (Feb 24), Qoder (Apr 30), GitHub Copilot (Jun 1) billing changes
• ➕ Added GitHub Copilot Max tier ($100/mo, $200 AI Credits) and Claude Haiku 4.5
• 🐛 Fixed stale Cursor / Qoder / Windsurf / GitHub Copilot pricing throughout

2026-05-18

• ✨ added github PR review tools

2026-04-12

• ✨ added a website for easy navigation

2026-04-11

• ✨ Initial release

• Quick Comparison
• Free LLM API Providers
  • Fully Free Providers
  • Providers with Trial Credits
• AI-Powered IDEs
  • IDEs with Pro-Grade Models
  • IDEs with Basic Models
• CLI Coding Tools
  • CLI Tools with Pro-Grade Models
  • CLI Tools with Basic Models
• API Providers for AI Coding Tools
• Paid Tiers Comparison
• Local Models
• free-coding-models CLI
• Additional 2026 AI Tools
  • Agentic Workflow Platforms
  • Data Visualization & Analysis
  • Creative & Multimedia Tools
  • Productivity & Research Tools
  • Vertical AI
  • Marketing & SEO Tools
  • Open Source & Local Tools
• 🏗️ Recommended Stacks
• ⚡ Realtime & Streaming APIs
• 🎙️ Speech Models
• 🎨 Image Generation Models
• 🎬 Video Generation APIs
• 🌐 AI Browser Automation
• 💾 Cheap Vector DB Hosting
• 🏛️ Common AI Architecture Patterns
• 💵 Model Price Comparison
• 🎯 Best Models by Use Case
• ⏱️ Rate Limit Comparison
• ✅ Commercial Use Summary
• 🧩 RAG Stack Tools
• 🔢 Best Free Embedding APIs
• 🖥️ AI Hosting & GPU Providers
• 📊 AI Evaluation Tools
• 📐 Structured Output Tools
• 🏷️ Legend
• Contributing
• License

Models achieving ≥60% on SWE-bench Verified / Pro or leading agentic benchmarks (Terminal-Bench 4.0, DeepSWE v1.1):

Note: [verify] indicates scores need verification from official sources. Always check current benchmarks before making decisions.

Ready-made combinations for different use cases. Copy-paste these configurations.

Total Cost: $0/month

Best for: Real-time apps, trading bots, live coding assistants

Total Cost: ~$10/month for pro-grade everything

Best for: Healthcare, legal, finance - any sensitive data

Best for: Autonomous research assistants, code review bots, data processing pipelines

Best for: Document QA, legal document analysis, knowledge bases

Limits: 20 RPM, 29 free models (262K context max, March 2026), models share quota

• Llama 3.3 70B ✅
• NEW: Nemotron 3 Super (262K context)
• NEW: MiniMax M2.5
• NEW: Devstral 2 (Apache 2.0)
• NEW: Gemma 3n family (mobile-optimized)
• qwen/qwen3.6-plus:free ✅
• Hermes 3 Llama 3.1 405B
• Llama 3.2 3B Instruct
• Mistral Small 3.1 24B
• Full list

Unified API gateway for 100+ LLMs. OpenAI and Anthropic SDK-compatible. China-friendly with Hong Kong direct access (100-300ms latency). No monthly fees, pay per token.

Limits: Not published | 1 free model

• GLM-4.7-Flash (200K context, 128K output, $0/M input, $0/M output)

Data is used for training when used outside UK/CH/EEA/EU.

Rate limits: Tier 1 (default): 250 RPD | Tier 2: Requires $250 spend + 30 days

Note: Data training outside UK/CH/EEA/EU still applies.

Phone number verification required. Models tend to be context window limited.

Limits: 1K credits signup, up to 5K total, 40 RPM (phone verify required)

• 46+ models including Llama 3.3 70B, Llama 4 Scout, Mistral Large, Qwen3 235B

Free tier requires opting into data training; phone verification required

Limits (per-model): 1 req/s, 500K tokens/min, 1B tokens/month

• Open and Proprietary Mistral models (Mistral Large 3, Small 3.1, etc.)

Self-hosted, open-source AI gateway for routing requests across 20+ providers through an OpenAI-compatible API.

• Models: 20+ provider integrations through your own keys
• Pricing: Apache-2.0; no hosted subscription required

AI gateway with curated models. Free models may use data for improvement.

• MiniMax M2.5 Free (S+, 80.2% SWE-bench)
• MiMo V2 Pro/Omni/Flash Free
• Nemotron 3 Super Free
• DeepSeek V4 Free
• Trinity Large Preview Free

Limits: 10,000 neurons/day

• @cf/aisingapore/gemma-sea-lion-v4-27b-it
• @cf/ibm-granite/granite-4.0-h-micro
• @cf/openai/gpt-oss-120b, @cf/openai/gpt-oss-20b
• @cf/qwen/qwen3-30b-a3b-fp8
• @cf/zai-org/glm-4.7-flash
• DeepSeek R1 Distill Qwen 32B
• Deepseek Coder 6.7B Base/Instruct (AWQ)
• Deepseek Math 7B Instruct
• Gemma 2B/3 12B/7B Instruct (LoRA)
• Hermes 2 Pro Mistral 7B
• Llama 2 7B/13B Chat (FP16/INT8/AWQ/LoRA)
• Llama 3 8B Instruct, Llama 3.1 8B Instruct (AWQ/FP8)
• Llama 3.2 1B/3B/11B Vision Instruct
• Llama 3.3 70B Instruct (FP8), Llama 4 Scout Instruct
• Mistral 7B Instruct v0.1/v0.2 (AWQ/LoRA)
• Mistral Small 3.1 24B Instruct
• Qwen 1.5 0.5B/1.8B/7B/14B Chat (AWQ)
• Qwen 2.5 Coder 32B Instruct, Qwen QwQ 32B
• Phi-2, SQLCoder 7B 2
• And more...

Full-featured integrated development environments with built-in AI assistance.

Model: Adaptive routing with Claude Sonnet 5.5, Claude Opus 5.5, and GPT-6.1 Sol (accessible via AI Gateway / custom config)

• Free tier (Hobby): Limited Agent requests + Limited Tab completions/month + 1-week Pro trial
• Free models: Cursor Small, DeepSeek V4, Gemini 3 Flash / 3.5 Flash (Limited access)
• Premium tiers required for manual model selections like Claude Opus 5.5, Claude Sonnet 5.5, and GPT-6.1 Sol (note: OpenAI restricts direct standard picker access for GPT-6 Astra)
• Credit-based billing since Jun 2025: each paid plan includes a credit pool equal to its price; Tab completions unlimited, Auto mode effectively unlimited, credits only deplete when you manually pick a premium model
• AI-powered code editor with autonomous coding capabilities
• Pro ($20/mo or $16/mo annually): $20/mo credit pool + Unlimited Tab completions + Auto mode
• Pro+ ($60/mo or $48/mo annually): $70/mo third-party model credits (discounted usage curve for $60 fee) + 3x Pro usage + Background Agents
• Ultra ($200/mo or $160/mo annually): $400/mo credit pool (20x Pro) + Priority access
• Teams Standard ($40/seat/mo or $32/seat/mo annually): Pro-equivalent per seat + Centralized billing + Usage analytics + SAML/OIDC SSO
• Teams Premium ($120/seat/mo): 5x usage bundled per seat + Priority support + Advanced team administration
• Enterprise (Custom): Everything in Teams + Pooled usage + SCIM + AI code tracking API + Audit logs
• Bugbot: Shifted from flat $40/seat/mo to per-run usage charge (~$1.00–$1.50 per review run triggered on commit pushes)

Pricing

Models: Doubao (natively in China), Claude 3.5/3.7 Sonnet & GPT-4o (international routing)

• New token-based pricing (effective Feb 24, 2026) — replaced the legacy "fast/slow request" model
• Free: Limited usage, 5,000 auto-completions/month, Standard queue
• Lite ($3/mo): $5 basic usage + bonus, Unlimited auto-completions
• Pro ($10/mo): $20 basic usage + bonus, Unlimited auto-completions, SOLO mode included, 10 concurrent cloud tasks
• Pro+ ($30/mo): $90 basic usage + bonus (4.5x Pro), 15 concurrent cloud tasks
• Ultra ($100/mo): $400 basic usage + bonus, Model early access, 20 concurrent cloud tasks
• 7-day free Pro trial (replaces the legacy $3 first-month deal)
• Annual: Pro $90/yr ($7.5/mo), Pro+ $270/yr ($22.5/mo), Ultra $900/yr (~$75/mo)
• On-Demand Usage: pay-as-you-go at API rates after basic + bonus usage is exhausted
• Migration bonus: $20 in dollar usage for current Pro users who manually switch (valid 90 days)

Pricing | Documentation

Models: OpenAI (GPT-6.1 Sol), Anthropic (Claude Sonnet 5.5, Claude Opus 5.5), Google, xAI

• New quota-based pricing (effective Mar 19, 2026) — replaced the legacy "prompt credits" model
• Daily + weekly usage allowance instead of monthly credit pool
• Existing paid subscribers are grandfathered at the old price but moved to the new quota system (with a free extra week to try it)
• Free ($0): Light quota + Unlimited Tab completions + 1 app deploy/day
• Pro ($20/mo): Standard quota + Full model access (Claude Opus 5.5, Claude Sonnet 5.5, GPT-6.1 Sol) + Purchase extra usage at API price
  • ~7-27 messages/day on Premium Plus models (Opus 5.5, GPT-6.1 Sol)
  • ~8-101 messages/day on Premium models (Sonnet 5.5, Gemini Pro)
• Max ($200/mo) — NEW Mar 2026: Heavy quota (~6x Pro) + Priority support
  • ~42-170 messages/day on Premium Plus models
  • ~291-1,190 messages/day on Lightweight models (Haiku, Flash)
• Teams ($40/user/mo): Standard quota per seat + Centralized billing + Admin dashboard + Priority support
• Enterprise ($60+/user/mo): Custom volume + SSO + Audit logs

Pricing | Pricing Announcement (Mar 18, 2026)

Models: Multi-agent (frontend/backend/testing agents)

• Agent-first IDE - new 2026 category
• Multiple specialized agents coordinate across codebase
• Free preview tier with high usage limits
• VS Code-based

Best for: Full-stack development with natural language direction

Models: Qwen3.6-Plus (71.2% SWE), Qwen-Coder-Qoder, GPT-6.1 Sol

• Free tier: Unlimited completions + limited chat/agent (basic models) + 2-week Pro trial (1,000 credits)
• Experts Mode: Multi-agent collaboration (new Mar 2026)
• Quest Mode: Fully autonomous app building
• Nextnew: Tab predictions
• Windows/macOS, VS Code-based
• Post-promotional standard pricing (settled after launch promo):

Pricing (standard, post-promo):

• Free: Basic models, limited messages
• Pro: $30/mo — 2,000 credits (or CNY 59/mo on Qoder CN All-in-One)
• Pro+: $60/mo — 6,000 credits
• Ultra: $200/mo — 20,000 credits
• Teams: $40/seat/mo — 3,000 credits/seat
• Personal Add-on Credits: $20 for 1,000 credits
• Credits: $0.02/credit, expire 1mo
• Teams new capabilities (rollout): BYOK, Security controls over MCP/Skills, Plugin management, Knowledge Engine

Docs | Pricing | Adjustment Notice

Models: Bring your own API keys (any provider)

• Open-source AI-powered coding assistant for VS Code
• Whole dev team of AI agents in your editor
• No subscription required - pay-as-you-go with your own keys
• Custom modes for different coding tasks

GitHub | Website

Model: Base model (Llama 3.3 70B), pro-grade models require subscription

• Individual plan: Free forever with unlimited code completions, AI chat, commands
• 70+ programming languages supported
• IDE integrations: VS Code, JetBrains, Vim/Neovim, Jupyter
• No credit card required
• Limited context awareness (expanded in paid tiers)
• Pro ($10/mo): Unlimited usage with advanced context awareness, Claude Sonnet 5.5, GPT-6 access
• Teams ($12/user/mo): Pro features + team management
• Enterprise (Custom): On-premise deployment, custom models

Pricing | Documentation

Models: Local models + cloud models with limited quota

• AI Free tier included with IDEs
• Unlimited code completion and local model support
• Limited quota for cloud-based features
• 30-day AI Pro trial included
• Offline mode with local models via Ollama/LM Studio
• AI Pro ($15/mo): Increased cloud quota + unlimited local models
• AI Ultimate ($25/mo): Maximum cloud quota + advanced features

AI Pricing | AI Features

Models: Claude Sonnet 5.5, GPT-6.1 Sol, Llama 3.3 70B, proprietary models

• Free tier with limited features
• Basic AI code completions and chat (limited)
• Local processing available
• Context heavily limited in free tier
• 600+ programming languages supported
• Pro ($12/mo): Enhanced AI completions and chat
• Enterprise ($39/user/mo): Multiple LLMs, private deployment, on-premises and air-gapped options

Pricing

Models: Unspecified models

• $1 credit/mo = ~100K tokens (reduced Mar 2026)
• Specific model not publicly specified
• Credit card required
• $20/mo: 20M tokens/month
• $200/mo: 200M tokens/month

Token Documentation

Models: Unspecified models

• 5 daily credits, max 30 per month (free)
• Models not publicly enumerated
• Credit card required
• Pro ($25/mo): 150 credits/month (5 daily credits)
• Teams ($30/mo): Higher limits (undisclosed)

Messaging Limits

Models: Proprietary models (not frontier)

• $5 in credits/month limit
• Uses proprietary models with varied routing
• Credit card required
• GPT-6 / Claude 5.5 access requires v0 Premium subscription

Updated Pricing Blog

General-purpose chat interfaces with free tiers.

Command-line tools for AI-assisted coding in your terminal.

Models: Gemini 3.5 Flash, Gemini 3.5 Flash-Lite, Gemini 3 Flash

• Gemini 3.5 Flash is the default high-throughput model (1,500 req/day on AI Studio free tier)
• Gemini 3.5 Flash-Lite provides ultra low-latency CLI completions
• 1,500 requests/day for Gemini 3.5 Flash / Gemini 3 Flash
• No credit card required for free tier
• MCP server support, Google Search grounding
• Install: npm install -g @google/gemini-cli

Rate Limits | Pricing

Models: Claude Sonnet 5.5, GPT-6.1 Sol, GPT-6 Astra (Ultra)

• Requires active paid Atlassian Cloud organization (Jira/Confluence)
• Organization quota: 5M tokens/day allocation for active subscriptions
• Token limits reset at midnight UTC
• Deep Jira & Confluence issue/doc integration and native MCP server support
• Pro ($19.99/user/mo): 100 tasks/day, 5x higher limits
• Ultra: 300 tasks/day, 20x higher limits, priority access to latest models (GPT-6 Astra)

Documentation | Token Limits

Models: GPT-6.1 Sol, Claude Sonnet 5.5, Gemini 3.5 Flash

• 150 AI credits/month (first 2 months), then 75 AI credits/month
• No credit card required for basic signup
• AI-powered terminal with code generation
• Build ($20/mo): 1,500 AI credits/month
• Bring Your Own API Key (BYOK) option available

Pricing

167k+ GitHub stars • 850+ contributors • 6.5M monthly users • Apache 2.0

Models: 75+ providers via BYOK — Anthropic, OpenAI, Google, Groq, AWS Bedrock, Azure, OpenRouter, local Ollama

• MIT/Apache 2.0 licensed — fork, customize, self-host
• Five agent modes (Tab-switchable): Build (full tools), Plan (read-only), Debug, Review, Docs
• LSP-driven self-correction — auto-spawns Language Server Protocol servers and feeds compiler diagnostics back to the model
• Multi-agent support: up to 10 parallel agents per workspace
• Local inference via Ollama: $0 — no data leaves your machine

OpenCode Go (recommended for getting started): Subscription bundle of curated open-weight models

• $5 first month, then $10/mo (beta)
• Models included: GLM-5.1, Kimi K2.5, MiniMax M2.5, DeepSeek V4 Pro/Flash, Qwen3.7 Max, MiMo-V2.5-Pro
• Usage limits: $12/5h, $30/week, $60/month
• "Use balance" option falls back to your Zen credits when limits are hit

OpenCode Zen: Pay-per-request credits (PAYG from $20)

Install: curl -fsSL https://opencode.ai/install | bash • brew install opencode • npm install -g opencode-ai

GitHub | OpenCode Go Docs

Models: Claude Sonnet 5.5, GPT-6.1 Sol, Gemini Flash, Grok Code Fast 1 (Free tier); Claude Opus 5.5 & GPT-6 Astra available in Pro/Pro+/Max/Business/Enterprise only

• MAJOR: Usage-based billing effective Jun 1, 2026 — premium request units (PRUs) replaced by GitHub AI Credits (token-based)
• 50 agent mode or chat requests + 2,000 completions/month (Free tier)
• Agent Mode with autonomous multi-step coding
• No credit card required for Free
• Free Copilot Pro for students/educators (GitHub Student Pack)
• Code completions and Next Edit suggestions remain included on all plans and do not consume AI Credits
• Pro ($10/mo): $15 monthly AI Credits + unlimited completions + cloud agent
• Pro+ ($39/mo): $70 monthly AI Credits + 1,500 premium req equivalent + Opus 5.5 & GPT-6 Astra access
• Max ($100/mo) — NEW Jun 2026: $200 monthly AI Credits + Priority access to new models + 2.9x Pro+ usage
• Business ($19/user/mo): 1,900 monthly AI Credits ($19) + unlimited completions (summer promo ended Sep 1, 2026)
• Enterprise ($39/user/mo): 3,900 monthly AI Credits ($39) + unlimited completions

Plans Details | Usage-Based Billing Announcement (Apr 27, 2026)

Model: Gemini 3.5 Flash / Gemini 3 Pro (upgraded from 2.5 Pro)

• 15 tasks/day free tier
• 3 concurrent tasks
• Rolling 24-hour window reset
• Pro ($19.99/mo): 100 tasks/day, 5x higher limits
• Ultra (via Google AI Ultra): 300 tasks/day, 20x higher limits, 60 concurrent tasks, priority access to latest models

Usage Limits | Documentation

Amazon's generative AI-powered assistant for software development (evolved from Amazon CodeWhisperer).

Models: Claude 3.5 Sonnet & AWS Bedrock foundation models

• Free Tier: 50 chat interactions and code transformations per month with an AWS Builder ID (no credit card required)
• Code completion, explanation, refactoring, and test generation in the IDE
• Supported IDEs: VS Code, JetBrains IDEs, Visual Studio, and AWS Cloud9
• Terminal integration via AWS CLI and macOS terminal
• Pro Tier ($19/user/mo): Higher usage quotas, enterprise user management, and custom code reference tracking

Pricing | Documentation

Xiaomi's subscription plan for AI coding scenarios — bundled access to MiMo flagship models Compatible with OpenCode, OpenClaw, Claude Code, and other mainstream toolchains

Models: MiMo-V2.5-Pro, MiMo-V2.5, MiMo-V2.5-TTS, MiMo-V2-Omni

• No context-length multiplier — same rate for 10K or 500K context (big deal for agentic workflows)
• 1:2 credit ratio for Pro vs Omni models (consumed in parallel, not independently)
• Night discount: 0.8x consumption (00:00–08:00 Beijing Time)

Monthly Pricing:

API Pricing (permanent schedule announced May 27, 2026):

Models: Claude Sonnet 5.5 (default high-velocity coding), Claude Opus 5.5 (architectural refactoring), Claude Haiku 5.5

• Free tier available with limited usage
• Pro ($20/mo): Sonnet 5.5 access with extended usage + Opus 5.5 access
• Max 5x ($100/mo): ~225 messages/5 hours
• Max 20x ($200/mo): ~900 messages/5 hours
• Extended thinking modes: "think" (~4K tokens), "megathink" (~10K), "ultrathink" (~32K)

Pricing

Models: GPT-6.1 Sol, GPT-6 Astra (with up to 8x faster token generation via Codex Ultrafast)

• Free with ChatGPT Plus ($20/mo): 30–150 messages/5 hours with GPT-6.1 Sol
• ChatGPT Pro ($200/mo): 300–1,500 messages/5 hours with GPT-6 Astra
• Pay-as-you-go API: $2.00/$10 per million tokens (GPT-6.1 Sol input/output)
• First model with session "compaction" for multi-million token deep sessions

GitHub Repo

Models: OpenAI- and Anthropic-compatible endpoints, Gemini, and local models

• Self-hosted QA CLI and dashboard with no agent-qa usage cap
• No credit card required when paired with local models
• Bring your own model/provider; provider costs and rate limits apply
• MCP server and skills for Claude Code and Codex
• Install: npm install -D agent-qa && npx agent-qa init

Documentation | License | Last verified: 2026-08-16

These services provide API access to coding-optimized models for tools like Cursor, Continue.dev, Cline, etc.

• 50 requests/day free tier (1,000/day with $10+ credits)
• Qwen3-Coder-480B, Qwen3-30B-A3B, Qwen3-235B-A22B, Gemini Flash
• OpenAI-compatible API

• 1.5M tokens/day free tier (expanded Feb 2026)
• 30 req/min, 8,192 token context
• Models: Qwen3.6-Plus-480B, Llama 3.1 70B
• Ultra-fast: 2,400 t/s (Qwen3.6)
• OpenAI-compatible API (works with Cursor, Continue.dev, Cline, RooCode, etc.)

Pricing

Running open-weight frontier models locally provides unlimited coding assistance without API costs.

Notable Local Models (2026):

• Qwen3.6-Plus-480B (71.2% SWE, ~150GB VRAM)
• Gemma 4 [verify] (Google, Apache 2.0, fully open-source flagship)
• GLM-5.1 / GLM-5V-Turbo [verify] (Zhipu MoE-based SOTA coders)
• Devstral 2 (24B, Apache 2.0, agent-optimized)
• DeepSeek Coder V4 (lite version ~18GB)

Find the fastest free coding model in seconds. Ping 271 models across 25 providers in real-time.

npm install -g free-coding-models
free-coding-models

• Parallel pings — all 271 models tested simultaneously
• Stability Score (0-100) — composite score from p95 latency, jitter, spike rate, uptime
• Smart ranking — top 3 highlighted 🥇🥈🥉
• Favorites — star models with F, persisted across sessions
• Tool Integration — auto-configure OpenCode, Goose, Aider, Continue, Cline, etc.
• OpenCode Zen Models — Curated free models (MiniMax M2.5 Free, MiMo V2, etc.)

# Most reliable model right now
free-coding-models --reliable

# Configure Goose with S-tier model
free-coding-models --goose --tier S

# NVIDIA top models only
free-coding-models --origin nvidia --tier S

# JSON output for scripting
free-coding-models --tier S --json | jq -r '.[0].modelId'

All 271 models allow commercial use of generated output. You own what the models generate.

Key Points:

1. Generated code is yours — no model claims ownership of your output
2. Apache 2.0 / MIT models (Qwen, GLM, GPT-OSS, MiMo, Devstral Small) are the most permissive — no strings attached
3. Llama requires "Built with Llama" attribution; >700M MAU needs a Meta license
4. DeepSeek / MiniMax have use-restriction policies (no military use) that govern the model, not your generated code
5. API-served models (Claude, Gemini, Perplexity) grant full output ownership under their terms of service

⚠️ Disclaimer: This is a summary, not legal advice. License terms can change. Always verify the current license on the model's official page before making legal decisions.

• Goal: Compare AI coding tools by their access to pro-grade models and free tier limits.
• What qualifies a model as "pro-grade"? Models must achieve ≥60% on SWE-bench Verified / Pro or leading agentic benchmarks (Terminal-Bench 4.0, DeepSWE v1.1), demonstrating real-world software engineering capability. Current qualifying October 2026 models: Claude Opus 5.5 (Flagship Enterprise reasoning, multi-hour sandboxing & prompt-injection defense), GPT-6 Astra (Autonomous Agentic Flagship / Computer Operator), GPT-6.1 Sol (matches Astra on DeepSWE v1.1, leads OSWorld 2.0 & Terminal-Bench Science), Claude Sonnet 5.5 (70.6% Terminal-Bench 4.0, 55.5% CursorBench 4.0, 30% fewer tokens/task), Gemini 4 Argon (Frontier Long-Horizon with 1M output tokens), Gemini 3.5 Flash (84.2% CharXiv), and Qwen3.6-Plus (71.2%).
• [verify] tag: Indicates information needs verification from official sources. Pricing, limits, and model availability change frequently.
• Different limit types: Tools use various quota systems - requests, tokens, credits, chats - making direct comparison challenging. Check documentation for specifics.
• Real-world usage: Actual consumption varies dramatically based on coding style, task complexity, and tool implementation.

Visual orchestration tools for building autonomous AI agents without coding.

AI-powered tools for conversational data analysis and narrative visualization.

Professional-grade content creation with generous free tiers.

Medical AI:

Legal AI:

Low-latency APIs for voice assistants, live coding copilots, trading tools, and realtime chat.

Best for:

• Trading bots: Groq streaming (fastest)
• Voice assistants: OpenAI Realtime API (end-to-end)
• Live captions: AssemblyAI or Deepgram
• Realtime chat: Gemini Live API

Speech-to-text and text-to-speech models comparison.

Comparison of image generation models and APIs.

Text-to-video and image-to-video generation. Hot area in 2026.

Tools for AI agents to control browsers - web scraping, form filling, testing.

Stack Recommendation:

• AI agents: Stagehand + Browserbase
• Web scraping: Playwright + Scrapy
• Testing: Playwright + AI assertions

Production-ready vector storage without high costs.

Recommendation by Stage:

• MVP: ChromaDB (local) → Supabase (hosted)
• Production: Qdrant Cloud or Pinecone
• Enterprise: Milvus or Weaviate

Proven patterns for building AI applications.

User → Chat UI → LLM API → Response
            ↓
        Context Memory (Redis/Postgres)

Stack:

• Frontend: Next.js + Vercel AI SDK
• Backend: FastAPI + OpenRouter
• Memory: Upstash Redis or Supabase

Documents → Chunking → Embeddings → Vector DB
                                    ↓
User Query → Embedding → Similarity Search → LLM → Response

Stack:

• Framework: LlamaIndex or LangChain
• Embeddings: BGE-Large or Jina v3
• Vector DB: ChromaDB (dev) → Pinecone (prod)
• LLM: Claude Sonnet 5.5 or GPT-6.1 Sol

User Request → Agent Controller → Tool 1 (Search)
                              → Tool 2 (Code exec)
                              → Tool 3 (API call)
                              ↓
                        Synthesize → Response

Stack:

• Framework: LangGraph, AutoGen, or CrewAI
• Tools: Function calling with Claude 5.5 / GPT-6.1
• Memory: Vector DB + State management
• Monitoring: LangSmith or Arize

User Request → Router (classify intent)
                    ↓
    ┌───────────────┼───────────────┐
    ↓               ↓               ↓
Cheap Model    Medium Model    Expensive Model
(GPT-6 Luna)     (Claude Sonnet 5.5)   (Claude Opus 5.5 / GPT-6 Astra)
    ↓               ↓               ↓
Simple Q&A    Complex task    Hard reasoning

Implementation:

• Router: Fine-tuned classifier or LLM-based
• Cost optimization: Route 80% to cheap models
• Fallback: Escalate if cheap model fails

Audio Input → STT → LLM → TTS → Audio Output
     ↓           ↓      ↓       ↓
 Deepgram    Groq   Claude  ElevenLabs

Stack:

• STT: Deepgram or Whisper Streaming
• LLM: Groq for speed or OpenAI Realtime
• TTS: ElevenLabs or OpenAI TTS
• Latency target: <500ms end-to-end

Image Input → Vision LLM → Structured Output
                                 ↓
                          Database / Action

Stack:

• Vision: GPT-6.1 Sol Vision, Claude Sonnet 5.5, or Gemini 3.5 Flash
• Structured output: Instructor + Pydantic
• Storage: Postgres JSONB or MongoDB

Text Prompt → LLM Enhancement → Image Gen → Upscaling
                                                ↓
                                           Video Gen (optional)

Stack:

• Enhancement: GPT-4 or Claude
• Image: FLUX or DALL-E 3
• Upscale: Upscayl or Magnific
• Video: Runway or Pika

API pricing for budget planning. Sorted by input cost.

💡 Pro tip: DeepSeek's 90% cache discount makes it cheapest for repetitive tasks with long prompts.

⚠️ October 2026 Generation Note: Claude Sonnet 5.5 lowered rates to $2.00/$10.00 ($0.20 cache read) with 30% lower token usage per task. GPT-6.1 Sol provides near-Astra capabilities at $2.00/$10.00 ($0.10 cache read). Claude Haiku 5.5 is rolling out to replace Haiku 4.5. Claude Opus 5.5 ($5/$25) and GPT-6 Astra ($10/$50) represent the top enterprise and autonomous tiers.

Don't just use SWE-bench - match models to your specific task.

Critical for scaling applications. Plan your architecture.

Quick reference for legal safety.

⚠️ Always verify current ToS - licenses can change.

Build document Q&A and semantic search systems.

Essential for RAG - don't overlook these.

Scale beyond free tiers.

Benchmark your models before production.

Force LLMs to return valid JSON/schemas.

Quick reference for badges used in this guide.

If you spot an error, missing source link, or have updated quota/model information, please open an issue or pull request with a source.

No affiliation with any vendor. All trademarks belong to their owners. Information is for research; accuracy not guaranteed; limits/pricing change frequently.

• cheahjs/free-llm-api-resources (18.4k ⭐) - Comprehensive free LLM API list
• mnfst/awesome-free-llm-apis (2.1k ⭐) - Permanent free LLM API tiers
• inmve/free-ai-coding (648 ⭐) - Pro-grade AI coding tools comparison
• Coding with AI - Practical techniques for coding with LLMs
• nowork-studio/awesome-ai-startups - A curated list of bootstrapped, pre-seed, and angel-funded AI products built by independent founders

This list was compiled and verified using:

• Gemini - For research and discovering new/additional AI tools
• Perplexity - For verifying information accuracy and checking if data is current
• Community repos - All referenced repositories above were used as reference sources

MIT © ShaikhWarsi

Last updated: October 5, 2026 • PRs/issues welcome