ThinkMaaS Free Token HunterFree quota, line by line
We run our own verified vendor accounts, use ThinSpider's crawling capability to keep discovering and tracking public free model capacity, meter every call through the ThinkMaaS gateway, and publish the source list and the limits line by line. Email-verified registration is open; the trial grant and daily check-in become claimable once the board opens, through the same OpenAI-compatible endpoint.
Free quota exists because of upstream vendors' free policies and changes with them — it can be narrowed, throttled or withdrawn. This page makes no promise of unlimited quota, permanent free access, or availability and service levels on free quota.
Free quota available today
The page embeds a static snapshot generated daily and sends no cross-origin live request; every number carries an observation time and a source
Sources: openrouter.ai/api/v1/models (counted from our runtime host, 2026-10-04); platform switch states from the public fields of maas.thinkalike.com.cn/api/status (2026-10-04). The machine-readable snapshot behind this board is the hub-snapshot block in the page source, the same source as the table below.
Every row carries a source and a verification date
Only entries reachable from our runtime host and verified are listed; unreachable or un-reviewed sources are in the Excluded table below
| Provider | Model ID | Context | Needs your own key | Rate limit | Verified | Source |
|---|---|---|---|---|---|---|
| apodex | apodex/apodex-1.1-mini:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| cohere | cohere/north-mini-code:free | 256000 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| dots-studio | dots-studio/dots-3-note-preview:free | 512000 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
google/gemma-4-26b-a4b-it:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models | |
google/gemma-4-31b-it:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models | |
| inclusionai | inclusionai/ling-3.0-flash-sante:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| liquid | liquid/lfm-2.5-2.6b:free | 65536 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| nvidia | nvidia/nemotron-3-nano-omni-30b-a3b-reasoning:free | 256000 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| nvidia | nvidia/nemotron-3-super-120b-a12b:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| nvidia | nvidia/nemotron-3-ultra-550b-a55b:free | 1000000 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| nvidia | nvidia/nemotron-3.5-content-safety:free | 128000 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| nvidia | nvidia/nemotron-3.5-lightning:free | 1000000 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| poolside | poolside/laguna-s-2.1:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| poolside | poolside/laguna-xs-2.1:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| qwen | qwen/qwen3.8-27b:free | 262144 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| thinkingmachines | thinkingmachines/inkling-small:free | 1048576 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
| thinkingmachines | thinkingmachines/inkling:free | 1048576 | Available at launch (not wired yet) | 厂商未公布限速 | 2026-10-05 | openrouter.ai/api/v1/models |
Rate-limit source: the OpenRouter Limits documentation (consulted 2026-10-04) — free-model requests are capped at 20 requests/minute, at 50 requests/day while cumulative spend is under 10 USD and 1000 requests/day once it reaches that. Which models are actually available follows the day's board.
Domestic free tiers: Zhipu open platform (open.bigmodel.cn), Moonshot (platform.moonshot.cn), SiliconFlow (api.siliconflow.cn), Alibaba Cloud Bailian DashScope (dashscope.aliyuncs.com) and DeepSeek (api.deepseek.com) all returned 401 from our runtime host on 2026-10-04 (the route works, only a key is missing). Their free-tier figures change often, so this page does not hard-code them; model-level rows are appended by the daily snapshot once a source joins the published set.
How we differ from account-farming tools
Most "free token" projects on GitHub do bulk registration, captcha solving and fingerprint resets. We list them as risk intelligence — we do not implement or advertise them
| Project | Class | License | Stars observed | What it actually does | Do we use it |
|---|---|---|---|---|---|
| freellmapi | L2 free-pool aggregation | MIT | 30.5k | Aggregates free tiers of many providers behind one OpenAI-compatible API, routes by availability and switches on rate limits; uses user-supplied keys | Same class built first-party (L2), code not imported |
| awesome-free-llm-apis | L2 catalog | CC0-1.0 | 9.1k | A continuously verified list of LLM APIs with durable free tiers, with CI checking links and entries | Same class built first-party (L2) |
| workbuddy-auto-signin | L1 own-account automation | MIT | 972 | Reads the local login session of your own account and completes the daily check-in and reward claims; registers no new accounts | Same class built first-party (L1) |
| Free-LLM-Collection | L2 catalog | No LICENSE | 609 | A list of free LLM APIs with a selection of models (documentation only) | Not imported (no license) |
| FreeLLM-API-KeyHub | L2 catalog | No LICENSE | 260 | Free API resources and limited-time calls across Chinese LLM vendors; states it ships no API keys | Not imported (no license) |
| free-ai-models | L2 catalog tracker | MIT | 188 | Refreshes a free-model list from public sources every 24 hours with GitHub Actions, including context and rate limits | Same class built first-party (L2) |
| Free_Token | L2 bring-your-own-key | No LICENSE | 34 | Aggregates free keys the user has applied for manually into an OpenAI-compatible entry point | Not imported (no license) |
| AutoRouterKey | L3 bulk registration | MIT | 20 | Browser automation that creates Outlook mailboxes and registers OpenRouter accounts in bulk to harvest keys | Forbidden zone — risk intelligence only |
| Kiro-account-manager | L3 fingerprint / machine-ID reset | AGPL-3.0 | 1.4k | Rotates an account pool, auto-refreshes tokens, modifies device identifiers and machine IDs to evade association bans | Forbidden zone — risk intelligence only |
| any-auto-register | L3 bulk registration + captcha solving | AGPL-3.0 | 3.3k | Registers accounts automatically for 13+ platforms using disposable mailboxes and third-party captcha-solving services | Forbidden zone — risk intelligence only |
Class labels: L1 = automation limited to official free quota on accounts we own and verify; L2 = discovery and aggregation of public free model catalogs and official free tiers; L3 = bulk registration, captcha bypass, fingerprint resets, stealing other people's sessions. Our implementation scope: first-party adapters for L1 / L2 only. There is no registration endpoint call, captcha-service domain or fingerprint-reset path anywhere in the code; L3 findings stay as threat and risk-control intelligence. That is a capability boundary — and the reason this program can run long term.
Where quota comes from, how it is metered, how often it refreshes
Three lawful collection paths, all auditable. We do not bypass any vendor anti-fraud control — that constraint is itself part of the product.
- C1 public free model catalogs: crawl models vendors and aggregators publish for free (OpenRouter
:free, domestic free tiers), diff for changes and add them to the pool. Fully automatic. - C2 official free quota on our own accounts: daily check-ins, coupons and free-tier allowances on accounts we already own and verified. Scheduling is automatic; credentials and authorisation are configured once by a human.
- C3 one-time vouchers for our own new accounts: apply for an official API key or new-user voucher for our new account — registration is done by a human with real details, the tooling only records and claims.
:free of 466 entries · observed 2026-10-04| Source | Status | Why excluded | Observed | How it was checked |
|---|---|---|---|---|
| Google Generative Language API (Gemini) | Unreachable | HTTPS requests from our runtime host return code 000 (no connection established), so it neither enters the runtime pool nor is advertised as a free source | 2026-10-04 | vendor site · runtime-host egress test |
| Mistral | Unreachable | Same as above, return code 000 | 2026-10-04 | vendor site · runtime-host egress test |
| ModelScope | Unreachable | Same as above, return code 000 | 2026-10-04 | vendor site · runtime-host egress test |
| Groq | 403 | Its model-list endpoint returns 403 from our runtime host; there is no usable free path | 2026-10-04 | vendor console · runtime-host egress test |
| Hugging Face | Unreachable | Measured 2026-10-05 from our runtime host: code 000 (TLS/SNI blocked); its serverless free tier is also account-billed and redistribution is unverified | 2026-10-05 | huggingface.co |
| DeepInfra | Checked: no free tier | Its model catalogue is public and credential-free; of 183 entries measured, zero are priced at 0, so there is no free quota to hunt. This row records the check only and is excluded from public totals | 2026-10-05 | deepinfra.com |
| 5 domestic free tiers (Zhipu / Moonshot / SiliconFlow / Bailian / DeepSeek) | ToS review pending | The route works, but whether each vendor's terms allow distributing free quota to third parties through a gateway must be evidenced clause by clause before board-listing; until then the source stays published=false and is used only as trial credit | 2026-10-04 | Each vendor's developer terms page · free-tier endpoints returned 401 |
| L3-style tooling (bulk registration / captcha solving / fingerprint reset) | Deliberately excluded | Fraud against vendors or anti-fraud bypass, plus unauditable third-party code; recorded as risk intelligence only — not implemented, not advertised, not integrated | 2026-10-04 | See the Open-source landscape table on this page |
| Public field | Value (observed 2026-10-04) | What it means |
|---|---|---|
register_enabled | true | Open registration is enabled |
email_verification | false | Email verification is not yet enabled; turning it on is the precondition for rolling out grants |
checkin_enabled | false | Daily check-in is not yet enabled; this page does not pre-announce any check-in figure |
turnstile_check | false | No overseas human-verification service is used; protection relies on email verification and domain restrictions |
quota_per_unit | 500000 | Quota-to-currency conversion unit used for display |
Verification address: maas.thinkalike.com.cn/api/status (read-only public fields, 2026-10-04). Refresh cadence: catalog discovery once a day, quota snapshots 1–2 times a day, check-in style claims follow each vendor's own rules (usually once a day). The snapshot rewrites this page and the English counterpart, then ships with the site deploy. Daily issuance is controlled by monitoring plus a circuit breaker (issuance stops on hitting the cap, resumes after it falls back); we do not call that a hardware-level hard cap. What counts toward the daily exposure: the registration grant, daily check-ins, and manual top-ups that land on a registered member of the free-pool group. Top-up rows whose target user cannot be resolved (e.g. our own acceptance script funding a throwaway account) are reported but not metered, When the free pool can open is not something we assert verbally: the gateway self-publishes a public read-only readiness JSON every 10 minutes (pool-readiness.json) listing exactly what is missing, and the board page shows that reason.
Five steps from signup to a working free call
The endpoint and the calling convention are identical to a paid deployment, so a trial doubles as an integration test
Verify email
Email verification is the precondition for rolling out grants; its current switch value is in the table above, and no grants are issued before it is on.
Get the grant
On the day the channel opens, the registration grant figure is written into this page's snapshot — it is not announced in advance.
Check in daily
Check in on the account page for that day's quota; when the daily cap is reached, issuance stops for that day and resumes the next.
Call the free pool
Point base_url at https://maas.thinkalike.com.cn/v1 and use a model ID from the day's free-pool board.
Copy and run: curl and the OpenAI SDK
curl https://maas.thinkalike.com.cn/v1/chat/completions \
-H "Authorization: Bearer $THINKMAAS_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen/qwen3.8-27b:free",
"messages": [{"role": "user", "content": "Introduce yourself in one sentence"}]
}'from openai import OpenAI
client = OpenAI(
base_url="https://maas.thinkalike.com.cn/v1",
api_key="YOUR_THINKMAAS_TOKEN",
)
resp = client.chat.completions.create(
model="qwen/qwen3.8-27b:free", # use a free-pool model ID from the day's board
messages=[{"role": "user", "content": "Introduce yourself in one sentence"}],
)
print(resp.choices[0].message.content)Note: issue your own token on the ThinkMaaS account page — this page never collects any credential. Model IDs must come from the day's board, because the free pool gains and loses entries with upstream policy. Free-quota users can only call models inside the free-pool group; paid channels are unavailable to them, and requests are throttled or stopped when quota runs out or the daily cap is hit.
Whether the quota is usable is decided by the people using it daily
This page is a daily snapshot; live observations, per-row sources and the comment area sit on the hunter board page, and comments are pre-moderated
How comments work
On the live board you can comment on a specific source or model: rate-limit behaviour, long-context quality, whether it has stopped working. Comments are pre-moderated — unapproved content never appears in the public list. Bodies are capped at 2,000 characters, escaped on output, outbound links always carry rel="nofollow", and every moderation decision is recorded with the reviewer.
Why it is worth writing
We are an aggregation platform: the same free quota pool is discovered, tracked and re-verified by a crawler engine. Your feedback is read next to the quota snapshot and decides whether a source stays listed, gets downgraded, or is removed — and removals are stated with a reason in the "excluded" table below rather than disappearing silently.
Go to the live board
The numbers on this page refresh once a day from a generated snapshot. For current observations, verification dates and the comment area, use the board.
Bring your teammates, hear about the board first
Invite rewards and board announcements both go through public configuration and this page's snapshot — no private promises
Invite for quota
The invite bonus is set by platform configuration (QuotaForInviter / QuotaForInvitee). Both are 0 today. There is a gate worth stating plainly: rc.40 refuses to write any invite reward until the platform compliance attestation (payment_setting.compliance_confirmed) is ticked, and that attestation can only be made by the platform operator in person — we will not click it on anyone's behalf. Once it is confirmed and enabled, the figure and rules land in this page's snapshot rather than in verbal promises. Your personal invite link is on the ThinkMaaS account page.
Request a ThinkMaaS trial
The free pool is for trying and validating; for stable capacity go to the paid gateway: one entry to all mainstream models, multi-channel orchestration, token quotas and cost dashboards, with air-gapped private deployment.
Operations group and broadcast
Source failures, pool changes and board announcements are pushed to our operations group; the group invite is provided by the account manager together with the trial approval. Prefer email? Write to contact@thinkalike.com.cn. This page's snapshot is itself the public broadcast — there is no separate subscription channel and no credential is ever requested.
About free quota, limits included
What is the Free Token Hunter?
A quota program on ThinkMaaS: ThinSpider keeps discovering public free model capacity and tracking its changes, we claim quota on our own verified vendor accounts, every call is metered through the ThinkMaaS gateway, and the source list and limits are published here.
Why is it free — and what do you get out of it?
Because these models are already offered for free by upstream vendors as free tiers or free pools; our cost is only the system that discovers, claims and meters them. Our motive is plain: a zero-friction entry point where teams test the OpenAI-compatible gateway and quota governance, which turns into long-term users and private-deployment customers. The free part comes from upstream policy, not from a subsidy promise by us.
Is the quota unlimited? Will it stay free?
No, and we make no such promise. Quota is bounded by upstream free policies and rate limits and can be narrowed, throttled or withdrawn at any time; on our side there is a daily issuance cap with a circuit breaker. You will not find the words "unlimited quota" or "permanently free" on this page, and we give no availability or service-level guarantee on free quota.
How do I claim it — can I claim it right now?
The path is: register → verify email → registration grant → daily check-in → call the free pool via https://maas.thinkalike.com.cn/v1. The authoritative switch state is the grant object embedded in this page's snapshot (register_enabled / email_verification / checkin_enabled, raw API values). While the snapshot reads not opened, the registration grant and daily check-in are off and no figures are announced in advance.
Why are Google Gemini, Mistral, ModelScope and Groq missing?
Measured from our runtime host on 2026-10-04: Gemini, Mistral and ModelScope return code 000 (unreachable) and Groq returns 403. Sources unreachable from our runtime do not enter the runtime pool and are not advertised; all of them are listed in the Excluded table with the reason.
Do you script bulk registrations to farm quota?
No, and this is a line we drew deliberately: bulk account registration (disposable email, SMS code services, bulk phone verification), captcha bypass or solving services, device-fingerprint / install-ID / machine-ID resets, IP rotation meant to evade vendor risk controls, and harvesting other people's cookies or tokens — none of it is done, written or integrated. Quota comes only from C1 public catalogs, C2 official free quota on accounts we own and verify, and C3 one-time vouchers we claim for our own new accounts.
Can I resell or commercially redistribute the free quota?
Not as use or as marketing. Whether upstream terms allow distributing free quota to third parties through a gateway has to be evidenced clause by clause; anything that has not cleared that review stays published=false and never enters the public board. Until review clears, free quota is used only under a trial credit framing.
Can free-pool users reach paid models?
No. Free-quota users are confined to the free-pool group; paid channels are invisible and unusable to them. Existing ThinKM gateway channels stay in a separate group and are unaffected by this program.
How often is the board updated, and why should I trust it?
Catalog discovery runs once a day, quota snapshots 1–2 times a day, check-in claims per vendor rules (usually daily). The page embeds a static snapshot and issues no cross-origin live request, so it cannot render blank; every number in it carries an observation time and a source link you can check line by line.
What if the quota runs out or I get throttled?
When the daily cap is hit or your grant is exhausted, requests are throttled or stopped — wait for the next day's check-in window, or move to a paid plan. For stable capacity see ThinkMaaS: smart multi-channel routing, automatic failover, token quotas, usage billing and a cost dashboard, with private deployment.
Run it free once, then decide whether to buy
Register, verify your email, claim the trial grant, call the free pool — one OpenAI-compatible path verified on the spot, with sources and limits published line by line.
Request a ThinkMaaS trial