Nous Portal model catalog: https://t.co/EDqbWGouAF. Use the qwen/qwen4-max:free entry, not the paid one. The promo runs about 5 days from Nov 18, and the exact end date hasn't been published.
QWEN4-MAX AT $0 PER TOKEN. ENDS AROUND MONDAY.
@NousResearch's Portal has Alibaba's flagship (1M context) listed as a free model. The promo with Inferact and the Qwen team started Nov 18.
What their model page shows:
→ Two Qwen4-Max entries. Only qwen/qwen4-max:free costs $0. The other bills $1.42 in / $4.27 out per 1M
→ No token cap or rate limit published
→ Card: the Portal doesn't say. One aggregator claims signup asks for one, so check before you count on free
→ No end date announced. 5 days from the 18th puts it around Monday, Nov 23
My math: an 800K-token repo dump on the paid entry costs ~$1.14 in input alone. On :free it's $0, so I'd spend the window on a whole-codebase review or Q&A over a long doc.
My tip: paste the :free model ID straight into your config instead of picking from a dropdown. The logging policy for free models isn't spelled out, so keep secrets out of your prompts.
Worth it if you have a big-context job ready this weekend. #LLM
~5,000 MINUTES OF AI VOICE CALLS. FREE.
@polyaivoice just raised the sign-up credit on its new voice model API, Dialog-RSN-1, from $20 to $100. No card needed.
The terms, from their pricing page:
→ $100 for accounts opened before Dec 1
→ No card = 2 calls at once. Add one and you get 10
→ Voice output is free until Dec 1. Text is billed as tokens
→ Phone numbers and orchestration aren't included. You bring your own stack
My math: it's 2¢ per connected minute, so $100 ≈ 5,000 minutes before token costs ($2/M in, $5/M out).
How I'd use it: sign up this week, build one narrow agent (booking a table, moving an appointment) and run as many test calls as you can before Dec 1, while voice output is still free. Keep the agent's replies short to save on output tokens.
Verdict: real credit from a company that's been around since 2017. The page doesn't say when the credit expires, so spend it early. #AI
@sunilmehta_si Agreed. They say inputs aren't stored or trained on, but that's a policy, not an audit. Evals and side projects are the sweet spot; client data stays elsewhere.
3,000 FREE GPT-OSS-120B CALLS. RESETS EVERY MONTH.
Found it on a Japanese dev blog: Sakura Internet's AI Engine, OpenAI-compatible, barely known in the US.
What their pricing page confirms:
→ 3,000 chat requests/mo on gpt-oss-120b, reset monthly
→ Card + phone on the Sakura Cloud account. The free plan doesn't bill and never auto-switches to paid: past the cap you get rate-limited (429)
→ Plus 10K embeddings, 50 Whisper, 50 TTS
→ Free signups are capped; it closes when full
Where it bites: the quota is per account. A public demo with 30 visitors × 100 messages eats the month.
How I'd use it: as an extra, not the engine. Anything that must work on call 3,001 (logins, pricing) stays in plain code. The model only writes summaries and friendlier error text.
Gotcha from that blog: copy the model ID from /v1/models; a guessed name errors out.
Verdict: a steady free pool, and they say inputs aren't stored or trained on. Worth a translated signup. #LLM
Repo (MIT): https://t.co/8cEEW8U5SH. Install: npm install -g @ludoai/cli (Node.js 20+, CLI 0.2.0+). The 32 toolbox tools run offline with no account. Only new art generation (paid plans, credits) costs money. Tip: set LUDO_TELEMETRY=0 to turn off usage telemetry.
Your coding agent can now measure its own hitboxes before you ever playtest.
@Ludo_AI open-sourced 32 free command-line tools (MIT) that Claude Code, Codex, Cursor, Copilot and Gemini CLI can call while they build your game.
The gap: agents write clean movement code, then guess the numbers. Wrong sprite scale, feet floating above the floor, a hitbox wider than the character.
What the toolbox handles, offline:
→ feet pivot and box/capsule colliders
→ sprite scale, spritesheets and tilesets
→ audio loop seams, plus Godot, Unity and Phaser exports
No account, no API key, no credits. Node 20+ and one npm install.
What I like: the agent stops eyeballing. It gets real measurements it can read back and fix, and that's where most AI-built prototypes feel off.
Catches: art generation in the same CLI is paid (skip "ludo login" and you stay at $0), and usage telemetry is on unless you set LUDO_TELEMETRY=0.
Repo below 👇 #gamedev
Google's announcement: https://t.co/jXPydnNQSO
You claim it inside AI Studio: one dialog creates the Developer Program profile and sets up billing. Each monthly credit expires 12 months after it's issued.
YOU PAY FOR GOOGLE AI PRO. THERE'S AN API BUDGET INSIDE IT.
$10 a month in Google Cloud credit for the Gemini API, claimable in @GoogleAIStudio since Nov 13. Subscribed for a few months? Credits are likely waiting.
What Google says:
→ $10/mo on Pro, $40 or $100 on Ultra tiers
→ Each month's credit lasts 12 months, so unused months stack
→ Card required: Cloud billing account + $5 minimum prepay (real money, kept as balance)
→ Primary subscriber only, no family sharing. Claiming creates your Developer Program profile
My tip: when you link billing, set a $1 budget alert. It doesn't cap spend, but you get an email when real charges start, not a surprise invoice.
Skip posts telling you to turn it into a proxy server. Fast way to break Cloud's terms.
Verdict: don't subscribe just for this. Already pay for Pro? That's API money you're leaving on the table. #LLM
Sakura AI Engine (site in Japanese, use your browser's translate): https://t.co/GbdJXEs9y6 (needs a Sakura ID + phone + card on a Sakura Cloud project; the free plan has a signup cap)
FREE $10 FOR CLAUDE OPUS 5.5. JUST A GITHUB LOGIN.
A Chinese API relay called ccroute just started giving it to every new account on V2EX. Know what's behind it before you plug it in.
What the post and the panel show:
→ $10 on signup, no card asked, no expiry listed
→ Opus 5.5 and Fable 5.2 on the price list (can't confirm what's served)
→ GitHub sign-in works. Email confirmations reportedly get lost
→ Invites: $15 to you, $5 to the friend
My math from its rate table: Opus output bills $20–26/M here, so the $10 is ~385K–500K output tokens. A few agent runs, not a month.
The catch: it claims "pure" Claude accounts behind it, which likely means subscriptions resold as an API. That can break Anthropic's terms and vanish overnight. 103-day-old domain, terms and privacy pages switched off, no named operator.
My tip: cap the key's quota, lock it to one model, check the usage log after your first call.
Fine for a throwaway eval. Never your work code. #LLM
@sirshibaninja@crustdata And no scraper to keep fixing when a site changes. Your agent just calls the API or the MCP server and gets structured people and company data back.
$250 OF FREE DATA API CREDIT FOR A UNIVERSITY EMAIL. NO EXPIRY.
@crustdata's student program gives $250 in people and company data, and every new account gets some free credits too. Not an LLM: it's the data your agent looks up, via API, CLI, Claude connector or MCP server.
How to claim:
→ Find your school on their page (top 50 in the US, about 60 listed incl. Oxford and NUS)
→ Copy its promo code, click "Have a promo code?" at sign-up, use your student email
→ Skip "Continue with Google": the code won't apply
→ Card: not mentioned on the student page
My tip: person search is 0.03 credits per result, a full enrich 1 to 9 credits per profile. Search wide, enrich only the shortlist, and $250 goes much further.
Real people's data: check privacy laws before you cold-email anyone. #buildinpublic
@crustdata Claim it here: https://t.co/IY7uZum8XE
Per the page, student credits don't expire. Student email not accepted? Their school pages say to write to [email protected]