{"version":"1.6","last_updated":"2026-10-20","last_synced_at":"2026-11-04T04:40:50Z","count":6,"announcements":[{"id":11,"title":"BAILU AI joins TierMux: 8 models free to run, with vision, 1M context, and a daily bonus","details":"BAILU AI (白鹿) is now in the catalog, and every model listed was verified with real requests -- all eight answer on a personal account with no card and no credit balance, carried by the daily bonus credit.\n\nThe picks:\n- bailu-apex-2.7 -- 512K context, reads images, exposes its reasoning, solid tool calling. The strongest all-rounder of the eight.\n- bailu-apex-2 -- 1M context for large repositories and long documents. One caveat: no structured tool calling, so use it for analysis and generation rather than agent loops.\n- bailu-2.8-lite -- 262K context, vision plus adjustable thinking depth\n- bailu-edge-thinking -- 200K context, chain-of-thought reasoning, and it reads images too\n- bailu-2.7-free -- 262K context, the 2.7 line BAILU sells as its production coding model\n- bailu-2.6-preview -- 256K context, tuned for speed while writing code\n- bailu-edge-2 -- 262K context, lightweight, made for cheap high-volume calls\n- bailu-auto -- BAILU picks the model for you from whatever it has capacity for\n\nGet an API key:\n1. Register at https://bailucode.com/api_platform/ -- email and password, or an email code.\n2. Open the console and create an API key.\n3. Claim the daily bonus credit while you are in there; it renews every day and is what keeps the billed models free to run.\n\nAdd it in the TierMux Agent: open the Providers tab, find BAILU AI, paste the key. Nothing else to configure -- the catalog is already loaded.\n\nThe base URL is https://bailucode.com/openapi/v1, and it answers both the OpenAI dialect (/chat/completions) and the Anthropic one (/v1/messages), so existing clients work unchanged.\n\nNot in the catalog, and why: bailu-turing needs an Enterprise API plan, and bailu-2.8, bailu-2.7 and bailu-2.7-1m currently answer \"temporarily unavailable\" on every attempt. They will be added if that changes."},{"id":15,"title":"3 free models via Vyce AI (GPT 5.6 Luna, Claude Sonnet 4.6, DeepSeek V4 Flash)","details":"Vyce AI is a new provider in the catalog -- a third-party OpenAI/Anthropic-compatible reseller proxy (not a model owner itself) fronting GPT, Claude, DeepSeek and a few other models.\n\n**What you need to activate it**\n1. An account at [vyceai.com](https://vyceai.com)\n2. **Verify with Discord or Google** in the dashboard -- every model returns \"verification_required\" until you do this, no matter your balance.\n3. An API key from the API Keys tab (format: sk-...)\n4. Point your client at the base URL https://vyceai.com/v1 -- OpenAI-compatible, so use the slug as the model id (Anthropic-style calls work too, at /v1/messages).\n\n**This is not a \"free model\" in the zero-price sense**\nEvery model on Vyce carries a real per-1M-token price, the free ones included: gpt-5.6-new $2 in / $2 out, claude-sonnet-4-6 $3 in / $15 out, deepseek-v4-flash $0.09 in / $0.18 out. What makes them free to you is a **Daily Rewards** credit you claim by hand in the dashboard: $10.00/day on the Free plan (Lite $25/day, Pro $30/day are paid upgrades of the same mechanic), resetting every 24 hours at midnight UTC. Nothing claims it automatically -- skip a day and you get nothing that day, and missing a day resets any streak bonus.\n\n**Which models are actually free**\nThe dashboard's own Models page sorts its 8-model catalog into a \"Free\" tab (3) and a \"Paid\" tab (5). Only the Free-tab three are in this catalog: gpt-5.6-new (270K context), claude-sonnet-4-6 (200K context), deepseek-v4-flash (128K context) -- all three support tool calls. The other five (nemotron-ultra-550b, nemotron-vision, deepseek-v4-flash-lr, and a $0.50/image Grok image endpoint among them) are excluded here on purpose; they draw on your balance, not the daily reward.\n\n**Worth knowing before you rely on it**\n- This is a small, referral-funded reseller with no public track record -- not Groq/Google/OVH-level established. Verification appears tied to Discord/Google signup incentives; funding and terms could change without notice.\n- gpt-5.6-new was showing as \"Degraded\" (~100+ second average latency) when this was checked -- expect it to be slower or occasionally unavailable, independent of the free/paid question.\n- The $10-30/day is an account-wide budget shared across whichever of the three models you call, not $10-30 per model."},{"id":17,"title":"Free gateway: 27 live models via DreamPrompting (Groq, Gemini, Mistral, NVIDIA, Cohere, Cloudflare, ...)","details":"DreamPrompting is a new provider in the catalog: a **free, OpenAI-compatible gateway** that resells the free tiers of around 27 upstream providers behind a single free account. It is a forwarding layer, not a model owner -- asking for a model can be answered by whichever upstream is live, and a request may **fail over to a different provider** on error or rate limit unless you set allow_fallback:false.\n\n**What you need to activate it**\n1. A free account and API key at [dreamprompting.com/login](https://dreamprompting.com/login) -- no card required.\n2. Base URL https://dreamprompting.com/api/v1 -- note the /api/v1 path (not the usual /v1).\n3. Model ids are route-qualified, e.g. groq/openai/gpt-oss-120b, google/gemini-2.5-flash, openrouter/openai/gpt-oss-20b:free. The leading segment pins one upstream; a bare model name (or \"auto\") lets the router pick and fail over.\n\n**The free tier has real, shared limits**\nNothing is billed per token, but usage draws on one **account-wide** quota shared across every model: 100 requests/minute burst per IP, 500,000 tokens per rolling 24h, and 5,000 requests per rolling 24h. Each request is also capped at 32,000 input tokens (larger returns HTTP 413) and 8,192 output tokens. Adding your own upstream provider key (BYOK) removes those requests from the shared quota.\n\n**What is in the catalog**\nThe 27 models that are live now are all free and tool-callable, with a mix of reasoning and vision: GPT-OSS 120B/20B (Groq), Qwen3.6 27B and Compound Mini (Groq), Mistral Small/Large and Nemo, Nemotron 3 Super/Nano (NVIDIA), Command R / R+ / A (Cohere), Gemini 3.1 Flash Lite and 2.5 Flash (Google, vision), plus OpenRouter, Cloudflare Workers AI, Hugging Face and Codestral routes. The provider lists about 30 more as \"coming soon\"; they are left out until they can actually answer.\n\n**Worth knowing before you rely on it**\n- This is a third-party aggregator built on other companies' free tiers -- upstream availability and the allowance itself can change without notice.\n- Gemini \"thinking\" is off by default (requests route to a lightweight Gemini); opt in per request with reasoning_effort or thinking:true.\n- The CDN in front of the API rejects generic default User-Agents with HTTP 403 (code 1010); send a descriptive one. The OpenAI SDKs, curl and Python requests are unaffected."},{"id":18,"title":"Longcat 2.5 Preview Free (2-Week Trial)","details":"Free, **1M-context** multimodal reasoning model by Meituan. Optimized for complex coding, agentic loops, and zero-retention privacy during its 2-week launch promotion."},{"id":19,"title":"7 new free providers, 45 models: ~450 requests/day + $5/month credit, VLM Run needs no key","details":"TierMux's catalog has **7 new free providers with 45 models**. **VLM Run works with no key at all.** The other six need a free account key from the provider.\n\n**Roughly how much you get (estimate)**\n- About **450 requests a day** under the published caps (Routeway 200, MegaNova 150, VLM Run 100), or about 13,500 a month.\n- That is roughly **10M tokens a day** if every request is a typical coding-agent turn (about 30K tokens in, 3K out; MegaNova's small 8K-16K context makes its turns shorter).\n- Plus Charm Hyper's **$5 credit every month**: about 23M tokens on DeepSeek V4 Flash, or about 1M on Kimi K3.\n- Plus three providers with no published daily cap (Typhoon, LLMTR, FreeInference), so the real total is higher.\n\n**Best picks**\n- **Charm Hyper**: 23 models, including frontier Kimi K3, Qwen3.8 Max and Qwen3.8 2.4T, plus DeepSeek V4 Pro/Flash, GLM-5.3 and MiniMax M3, all paid from the free monthly credit (100 Hypercredits, about $5).\n- **VLM Run (no key)**: Qwen3.8 27B with 168K context, tool calls and vision.\n- **LLMTR**: Space Bunny (a frontier stealth model, free until 30 Sept), Qwen3.8 27B, Dots 3 Note Preview, Agnes 3.0 Flash and Nemotron 3 Ultra.\n- **FreeInference** (Harvard): GLM-5.3 Flash, MiniMax M3, DeepSeek V4 Flash and Qwen3.6 35B.\n- **Routeway**: DeepSeek V4 Flash, MiniMax M2.7 and Glimmer 30B.\n- **MegaNova**: Manta Mini, Manta Flash and Mistral Small 3.2.\n- **Typhoon**: Typhoon 2.5 30B.\n\n**Quota per provider**\n| Provider | Free quota | Key |\n|---|---|---|\n| Charm Hyper | $5 credit a month across all models | Yes |\n| Routeway | 200 requests a day across its free models, 5 per minute | Yes |\n| MegaNova | 50 requests a day per model (150 total), 60 per minute | Yes |\n| VLM Run | 100 requests a day per IP, 10 per minute, 30 per hour | **No** |\n| Typhoon | 200 requests per minute per model, no daily cap published | Yes |\n| LLMTR | A daily quota per model; the size is not published | Yes |\n| FreeInference | Described as \"generous\"; the size is not published | Yes |\n\n**Before you use them**\n- **FreeInference logs every prompt and response**, and may open-source anonymized data. Keep private code away from it.\n- Typhoon collects usage data to improve its models.\n- Charm's credit runs out fast on frontier models: about 30 agent turns a month on Kimi K3, against about 700 on DeepSeek V4 Flash.\n- Routeway's free DeepSeek and MiniMax models have a 42K context window.\n- VLM Run's limit is per IP address, so everyone behind one network shares it."},{"id":20,"title":"24 new free models since Sept 28: MiniMax M3.1 Flash, Beam 1.0, K2 Horizon, Qwen3.8 Max. Charm Hyper is no longer free","details":"Since the last update on **Sept 28**, the catalog gained **24 free models** across 15 providers. Every one was checked against what a **free account** can call. Many are launch promos, so the **Free until** column matters.\n\n**Best picks for coding agents**\n- **MiniMax M3.1 Flash** (strong, 1M context, tools, vision). Works on **Kilo with no key**, plus OpenRouter, OpenCode and Ollama.\n- **Beam 1.0** by Reflection (strong, 501B, 1M context). On **Kilo with no key**, plus BazaarLink, Ollama, OpenCode, Nous and LLMTR.\n- **K2 Horizon 375B** by IFM (strong, 512K context). A reasoning model built for tool-calling work, free on LLMTR until Nov 5.\n- **Qwen3.8 Max** (frontier) and **Qwen3.8 Flash** on AIHubMix, while GMI Cloud's promo lasts.\n\n**All new models**\n| Model | Tier | Free on | Free until |\n|---|---|---|---|\n| MiniMax M3.1 Flash | strong | Kilo (no key), OpenRouter, OpenCode, Ollama | Ollama: Oct 22, 23:59 UTC. Others: launch offer, no end date published |\n| Beam 1.0 | strong | Kilo (no key), BazaarLink, Ollama, OpenCode, Nous, LLMTR | LLMTR: Oct 22, 23:59 Turkey time. Ollama and OpenCode: \"limited time\", no date |\n| K2 Horizon 375B | strong | LLMTR | Nov 5, 23:59 Turkey time |\n| Qwen3.8 Max | frontier | AIHubMix | 30 days from launch (about Nov 14), or until the promo credits run out |\n| Qwen3.8 Flash | strong | AIHubMix | Same promo as Qwen3.8 Max |\n| DeepSeek V4.1 Flash 1017 | strong | ModelScope | No end date; spends ModelScope's daily free credits |\n| DeepSeek V4.1 Flash | strong | OpenCode | Limited time, no date |\n| GLM 5.3 Flash | strong | OpenCode, AIHubMix | No end date |\n| Step 5 Preview | strong | NVIDIA, ModelScope | No end date |\n| LongCat 2.5 Preview | strong | xKiro | No end date |\n| Big Pickle | strong | xKiro | No end date |\n| Space Bunny Alpha | frontier | xKiro | No end date. Reported to be MiniMax M3.1 Flash before launch |\n| Mistral Large 4 | strong | xKiro | No end date |\n| Ling 3.1 Flash | strong | Requesty | No end date |\n| Nemotron 3.5 Super VL (vision) | mid | NVIDIA, OpenRouter, Kilo (no key), OpenCode, LLMTR, NagaAI | OpenCode: limited time. Others: no end date |\n| Nemotron 3.5 Lightning | mid | BazaarLink, NaraRouter | No end date |\n| Xing4.0 29B | mid | OpenRouter, Kilo (no key) | No end date |\n| Cat Jam (stealth) | mid | BazaarLink | No end date |\n| Glyph Cluster (stealth) | mid | BazaarLink, Kilo (no key) | No end date |\n| Apodex 1.1 Mini | mid | NagaAI | No end date |\n| LongCat 2.0 | mid | xKiro | No end date |\n| Qwen3.8 27B (Heabsy) | mid | LLMTR | Nov 2, 00:00 Turkey time |\n| Nemotron 3 Nano 30B | mid | LLM7 | No end date |\n| Solar Mini 4 | small | Nous | No end date |\n\n**Ending soon**\n- **Oct 22:** MiniMax M3.1 Flash on Ollama, and Beam 1.0 on LLMTR.\n- **Oct 31:** Poolside Laguna S 2.1 and XS 2.1 on OpenRouter, Kilo and Nous.\n- **Nov 2 to Nov 5:** Qwen3.8 27B (Heabsy) and K2 Horizon on LLMTR.\n- **About Nov 14:** Qwen3.8 Max and Flash on AIHubMix.\nTierMux switches each one off on its end date, so your routing never lands on an expired model.\n\n**Free-account limits to know**\n- **Kilo:** 200 requests an hour per IP, shared by every free model. No key needed.\n- **OpenRouter:** 20 a minute, 50 a day (1,000 a day once you have bought $10 of credit).\n- **BazaarLink:** 10 a minute and 60 weighted units a day, shared by its free models. No card.\n- **AIHubMix:** 5 a minute and 100 a day per model.\n- **NagaAI:** 10 a minute and 100 a day, shared.\n- **xKiro:** 500K tokens a day.\n- **Ollama:** 1 request at a time.\n- **ModelScope:** 200 free credits a day after login.\n- **OpenCode:** its free models only answer through OpenCode itself.\n\n**Charm Hyper is no longer free**\nOur Sept 28 update said Charm Hyper gave $5 of credit every month. **That is no longer true.** On Oct 16 Charm replaced its free plan with prepaid Pay As You Go, starting at $5. A free account can no longer call any Hyper model, so **all 25 Charm Hyper models are now off** in TierMux. If you still have Hyper credit, add the models back yourself."}]}