free-coding-models

Find the fastest free coding model in seconds.
Live latency, stability and verdicts for 253 models from 25 free AI providers, then install the one you pick straight into your favorite coding tool.

Works with: OpenCode CLI / Desktop / WebUI, OpenClaw, Crush, Goose, Aider, Kilo CLI, Qwen Code, OpenHands, Amp, Hermes, Continue, Cline, Xcode, Pi, ZCode, ForgeCode, Copilot, jcode, Caveman Code and more.

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

_{Then create a free account on one of the 25 providers to grab an API key.}

_{Made with ❤️ and ☕ by Vanessa Depraute (aka Vava-Nessa)}

---

💡 Why this tool?

There is a large catalog of free and free-limited coding models (25 providers / 253 live models, generated from sources.js). Which one is fastest right now? Which one is actually stable, versus just lucky on the last ping?

free-coding-models (FCM) answers that by pinging every model in parallel, showing live latency, and computing a live Stability Score (0-100) combining p95 latency, jitter, spike rate and uptime. Average latency alone is misleading: a model that randomly spikes to 6 seconds is not reliable.

Once you pick a model, FCM writes it directly into your coding tool's config and opens the tool, so you go from "which model?" to "coding" in under 10 seconds.

Other highlights: AI Speed Test benchmarks (real completions, AI Latency + TPS), Smart Recommend (3-question wizard), a Ctrl+P Command Palette, a persistent 24h probe cache shared by every surface, live quota from response headers, real-world telemetry scores, models.dev enrichment with drift detection, and a tier scale based on SWE-bench Verified (S+ ≥ 70% down to C). Details: docs/stability.md.

---

⚡ Quick Start

① Install (Node.js 18+, no native build, never needs sudo):

npm install -g free-coding-models
free-coding-models --help   # prints every flag

② Grab one free API key. One is enough to start; add more later with P inside the app:

• Groq → console.groq.com/keys
• Cerebras → cloud.cerebras.ai (lowest latency in the catalog)
• NVIDIA NIM → build.nvidia.com (biggest no-credit-card quota)

③ Launch & paste your key: free-coding-models. First run prompts for keys (Enter skips), models ping in parallel, rows light up green ✅.

④ Pick a model & launch your tool: ↑↓ navigate, Enter to write the model into your tool's config and launch it. Pre-target another tool from the CLI, or cycle live with Z:

free-coding-models --goose --tier S      # Goose, pre-filtered to S-tier only
free-coding-models --crush --origin groq # Crush, Groq models only
free-coding-models --fiable              # print the single most reliable model and exit

⑤ Go further:

• 🌐 Prefer a browser? free-coding-models web opens the Web Dashboard.
• 🔀 Want one endpoint that never dies? free-coding-models --daemon-bg starts the Smart Model Router.
• 🤖 Live inside an agent? Install the OpenCode plugin or the Pi extension.

---

🟢 Providers

25 active providers / 253 live models, sorted by live model count. Top 8:

🧾 What "free" means here: free is a property of the (provider, model) pair, never of the provider as a whole. A row is listed only when that exact model id costs $0 to call through that provider (permanent free tier, :free variant, or free plan), verified live at audit time. The same open-weights model can be free on one host and paid on another - paid siblings are deliberately excluded. Full breakdown and badge legend: docs/providers.md.

📖 Full table, free-tier limits, env vars, tier scale and provider notes: docs/providers.md (generated from sources.js by node scripts/generate-provider-table.mjs, so counts cannot drift). OpenCode Zen's free models are listed there too.

⚠️ Health probes consume provider quota: FCM auto-pauses a provider on 429, backs off exponentially per failing model, and shows a footer chip while a provider rests. Leave a key empty and anonymous liveness probes still work for most providers.

---

🎛️ The Terminal UI (TUI)

The TUI is the heart of FCM: a live, sortable table of every model with real latency, stability, verdict, and a one-key launch into your coding tool. Sorts are one letter (R rank, S SWE, V verdict, B stability...), filters are T (tier cycle), D (provider cycle), E (visibility cycle), and the essentials are:

Mouse support: click headers to sort, double-click a row to launch, right-click to favorite, scroll to navigate.

Languages

The TUI and Web Dashboard support English (default) and Simplified Chinese (zh-CN). Change Language in Settings; the preference is shared and saved in your local config.

📖 Full key table, mouse reference and workflows: docs/tui.md · Stability score and columns: docs/stability.md

---

🌐 The Web Dashboard

A real-time browser control center for the same catalog: full filter/sort parity with the TUI, per-row benchmarks, favorites, Smart Recommend, a Router Dashboard, token usage analytics, and shareable URL deep-links (?tier=S+&sort=verdict&origin=groq). Start it with free-coding-models web (port 3333) or free-coding-models --daemon (dashboard + router API on port 19280). Runs headless in Docker (ghcr.io/vava-nessa/free-coding-models).

📖 Modes, Docker, Compose and all env vars (FCM_HOST, FCM_PORT, FCM_WEB_PORT, FCM_ROUTER_TOKEN, FCM_ALLOWED_ORIGINS, FCM_DISCORD_*_WEBHOOK): docs/web-dashboard.md

---

🔀 The Smart Model Router

A local OpenAI-compatible daemon: point any coding tool at http://localhost:19280/v1 with model fcm (key fcm-local) and FCM routes every request to the best available model in your active set. Adaptive health probes, per-model circuit breakers, family-preserving failover, auto-heal of broken sets, and a Playground chat to try it without configuring a tool.

free-coding-models --daemon-bg      # start
free-coding-models --daemon-status  # inspect
free-coding-models --daemon-stop    # stop

📖 Full guide (probes, circuit breaker, failover, Playground, --sync-set, REST API): docs/router.md

🧪 Router v2 engine (built in)

The router daemon now runs the hardened v2 engine internally: content-validated failover (a 200 with empty or garbage output is a failure and fails over), quota-aware routing with Retry-After pauses, persisted circuit breakers with a DEGRADED warning state, per-request decision traces, an Anthropic /v1/messages endpoint, and "test via router" actions that exercise the real routing chain. Same command, same port, same endpoints - existing setups upgrade by updating the package.

Open the TUI with Shift+V (or the "Router v2" entry in the Web Dashboard) to see the live fallback chain, per-request attempt chains and run pinned-model tests (Ctrl+T). Docs: docs/router-v2.md.

---

📖 CLI Flags

Flags combine freely in any order. The most common:

Tool launchers (21): --opencode (default), --openclaw, --crush, --goose, --aider, --kilo, --qwen, --openhands, --amp, --pi, --hermes, --continue, --cline, --xcode, --copilot, --forgecode, --zcode, --jcode, --caveman, plus OpenCode Desktop / WebUI variants.

📖 Canonical full flag reference (every flag, verified against the parser): docs/flags.md · Tool-to-config mapping: docs/integrations.md

---

🔌 Agent Extensions

• OpenCode plugin (fcm-opencode, beta): /fcm scans and ranks, /fcm 1 switches models, /fcm rescan forces a fresh scan. Install: packages/fcm-opencode/README.md
• Pi extension (fcm-pi, beta): silent by default, /fcm re-scans and lets you pick, /fcm-list shows a ranked table, error-triggered picker on 4xx/5xx. Install: packages/fcm-pi/README.md
• Shared core (fcm-agent-core): one scan/rank/cache engine for both adapters, with a cross-tool cache and daemon integration. API: packages/fcm-agent-core/README.md

---

📋 Contributing

Issues, PRs and new provider integrations are welcome. To add a provider, see the format in sources.js and re-run node scripts/generate-provider-table.mjs so docs/providers.md stays in sync. Latency numbers are real round-trip times from your machine.

→ Development guide · Config reference · Sync-set · Contributors

---

⚖️ Model Licensing

You own the generated output: the code and text these models produce is yours to use commercially under current provider terms. The licenses below govern the model weights, not your output: Apache 2.0 (Qwen, GPT-OSS, Devstral, Gemma), MIT / permissive (GLM Flash, MiniMax), Modified MIT (Kimi K2), Llama Community License (attribution required), DeepSeek and NVIDIA Nemotron licenses, and hosted-API terms (Gemini, Mistral, OpenRouter-hosted). Verify on the model's official page before making legal decisions; this is a summary, not legal advice.

---

📊 Telemetry

FCM collects anonymous usage telemetry (app version, tool mode, OS, terminal family, a random install ID). No personal information, API keys, prompts, source code, or file paths ever leave your machine. Disable with --no-telemetry or FREE_CODING_MODELS_TELEMETRY=0.

---

🛡️ Security & Trust

Keys live locally in ~/.free-coding-models.json (0600) and are only ever sent to the matching provider endpoint. FCM never requires sudo and never executes arbitrary remote code. To report a vulnerability, see SECURITY.md.

---

Star History

---

Contributors

Created and maintained by Vanessa Depraute (@vava-nessa), with contributions from @erwinh22, @whit3rabbit, @skylaweber, @PhucTruong-ctrl, @chindris-mihai-alexandru, @serajbaltu, @stgreenb, @MoriDanWork, @fan92rus, @Muhammad95959, @FaintFlower, @lehneres, @ia-S-on, @bangla24bdrang-lab and @Sy57em1.

→ Full credits and highlighted contributions: CONTRIBUTORS.md

---

📚 More docs

Providers · TUI · Web Dashboard · Router · Flags · Integrations · Stability & columns · Sync-set · Config · Development · Other free AI resources

License

MIT