Ascent

Free AI keys for translation

Which AI services let you translate for free, how much, and how to get a key and use it in Ascent: ranked with the most generous free tiers first.

Checked on 4 November 2026 on the services' own pages, where they publish their limits. Free tiers change often: the service's own page is the last word. Where a figure comes from elsewhere, it says so.

The services, ranked

#ServiceFree tierCard needed?After the free tier
1Google GeminiFree on Flash and Flash-Lite models; daily limits shown in AI StudioNoGemini 3.6 Flash $0.75 in / $3.75 out per million tokens; Flash-Lite from $0.25 / $1.50
2GroqAbout 1,000 requests a day, 30 a minute, 200,000 tokens a day (gpt-oss-120b)No$0.15 / $0.60
3OpenRouterFree models: 50 requests a day, 1,000 after a one-time $10 top-up; 20 a minuteNoPer model; free models stay free
4MistralFree Experiment plan; limits shown in its consoleNo (a phone number)Per model
5NVIDIA NIMFree for developers; about 40 requests a minute (NVIDIA's forum)NoNot sold per token
6Z.ai (GLM)GLM-4.7-Flash and GLM-4.5-Flash free; limits not publishedNot statedGLM-4.7-FlashX $0.07 / $0.40
7Cloudflare Workers AI10,000 “neurons” a day, shared by all modelsNot statedLlama 3.3 70B $0.29 / $2.25
8CohereTrial key: 1,000 requests a month, 20 a minute; not for commercial useNoPaid plans
9SambaNova20 requests a day, 200,000 tokens a dayNo Paid tier once a card is added
10DeepL API Free500,000 characters a monthYes, not chargedDeepL API Pro
11Cerebras$5 of credit for 30 days; up to 1 million tokens a dayYesPay as you go
12DeepSeekNoneTop-up$0.15 / $0.60 off-peak, twice that at peak hours
13Alibaba Qwen1 million tokens a model for 90 days, on some models, for new accountsNot statedQwen3.8 Flash $0.15 / $0.47
14SiliconFlowSome small models free (not confirmed on its own pages)Not statedPer model
15Together AINoneYes, $5 to startQwen3.8 Flash $0.15 / $0.47
16Fireworks AINone statedPrepaid creditPer model
17Hugging FaceNone on a free account; $2 a month with PROCreditThe provider's own price
18OpenAI (ChatGPT)NoneYes, from $5gpt-4o-mini $0.15 / $0.60
19Anthropic (Claude)$5 of trial credit with a verified phone numberNo, for the trialClaude Haiku 5.5 $0.10 / $0.50
20xAI (Grok)NonePrepaid creditgrok-4.3 $1.25 / $2.50
21Moonshot (Kimi)None$1 to startPer model
–Your own serverFree: Ollama or LM Studio on your computerNoYour electricity

Prices are US dollars per million tokens, in / out. A comic page of speech balloons is a few hundred tokens each way.

Add the key to Ascent

  1. Open the Translate sheet: the translate button in the bottom bar.
  2. Under Translate with, choose My AI key, then Add a key (or Change).
  3. Tap Service and choose one from the list. Ascent fills in its Base URL and a Model.
  4. Tap Get a free key to open the service's key page in a new tab, make the key there and copy it. Come back to the Translate sheet: the service you chose is still chosen.
  5. Paste the key under API key, then tap Save and Translate.

In older versions of Ascent, choose OpenAI-compatible under Provider for any service below other than Google, OpenAI, Anthropic and DeepL, and type the base URL and model given for it.

What the service sees

The text you translate — the page's text and the words read from its pictures — goes to the service you chose, with your key. Its terms and privacy policy apply; some free tiers may use what you send to improve their models (Google says so of Gemini's free tier). Your key stays on your phone, stored encrypted, and is sent only to the service's address. To keep everything on the phone, choose On device under Translate with instead.

Step by step, service by service

1. Google Gemini (AI Studio)

  1. Open aistudio.google.com/apikey and sign in with a Google account.
  2. Tap Create API key, choose or create a project, and copy the key.
  3. In Ascent, choose Google (Gemini) and paste it. The model is gemini-3.6-flash; gemini-3.5-flash-lite is cheaper and usually allows more requests a day.

Free with no card. Google does not publish the free limits on its pages any more: AI Studio shows yours for each model. Text sent on the free tier may be used to improve Google's products; on the paid tier it is not.

2. Groq

  1. Open console.groq.com/keys and sign in.
  2. Tap Create API Key, name it, and copy it: it is shown once.
  3. In Ascent, choose Groq and paste it. Base URL https://api.groq.com/openai/v1, model openai/gpt-oss-120b (or openai/gpt-oss-20b, faster).

Groq's rate-limit page gives 30 requests a minute, 1,000 a day and 200,000 tokens a day for these models; adding a card raises them. llama-3.3-70b-versatile, which Ascent suggested before, is now sold to businesses only, so Ascent moves a saved Groq key to openai/gpt-oss-120b.

3. OpenRouter

  1. Open openrouter.ai/settings/keys and sign in.
  2. Tap Create Key and copy it.
  3. In Ascent, choose OpenRouter and paste it. Base URL https://openrouter.ai/api/v1, model openrouter/free, which sends each request to one of the free models of the day.

Free models (their names end in :free) allow 20 requests a minute and 50 a day; 1,000 a day once you have bought $10 of credit, which you keep for paid models. To pick one free model yourself, type its name, ending in :free, as the model.

4. Mistral

  1. Sign up at console.mistral.ai, verify a phone number and choose the free Experiment plan.
  2. Open API keys, tap Create new key and copy it.
  3. In Ascent, choose Mistral and paste it. Base URL https://api.mistral.ai/v1, model mistral-small-latest.

No card. Mistral shows the plan's limits in its console only; others report about one request a second.

5. NVIDIA NIM

  1. Open build.nvidia.com, open a model and tap Get API Key.
  2. Sign in or make an NVIDIA account with your email (this joins the free developer program), then copy the key, which starts with nvapi-.
  3. In Ascent, choose NVIDIA NIM and paste it. Base URL https://integrate.api.nvidia.com/v1, model meta/llama-3.3-70b-instruct, or another model from build.nvidia.com.

NVIDIA does not publish the free limits; its forum gives about 40 requests a minute.

6. Z.ai (GLM)

  1. Sign up at z.ai.
  2. Open API keys, create a key and copy it.
  3. In Ascent, choose Z.ai (GLM) and paste it. Base URL https://api.z.ai/api/paas/v4, model glm-4.7-flash (free) or glm-4.5-flash (free).

Z.ai's price list shows both Flash models as free; it does not say how many requests they allow. They think before they answer, so they are slower than the others. In mainland China, the same models are at https://open.bigmodel.cn/api/paas/v4, with a key from bigmodel.cn.

7. Cloudflare Workers AI

  1. Sign in at dash.cloudflare.com and copy your Account ID (on the account's overview page).
  2. Open API Tokens, tap Create Token, use the Workers AI template and copy the token.
  3. In Ascent, choose Cloudflare Workers AI, paste the token, and in the base URL put your account ID in place of ACCOUNT_ID: https://api.cloudflare.com/client/v4/accounts/ACCOUNT_ID/ai/v1. The model is @cf/meta/llama-3.3-70b-instruct-fp8-fast.

10,000 neurons a day are free, for all models together; how many pages that is depends on the model.

8. Cohere

  1. Sign up at dashboard.cohere.com: a trial key is made for you.
  2. Open API keys and copy the trial key.
  3. In Ascent, choose Cohere and paste it. Base URL https://api.cohere.ai/compatibility/v1, model command-a-plus-05-2026.

A trial key allows 1,000 requests a month and 20 a minute, and may not be used commercially.

9. SambaNova

  1. Sign up at cloud.sambanova.ai.
  2. Open APIs, generate a key and copy it: it is shown once.
  3. In Ascent, choose SambaNova and paste it. Base URL https://api.sambanova.ai/v1, model DeepSeek-V3.1.

Without a card: 20 requests and 200,000 tokens a day. Enough to try, not for a long series.

10. DeepL API Free

  1. Sign up for DeepL API Free at deepl.com/pro-api. It is a separate account from the DeepL Translator app, and asks for a card, which the free plan does not charge.
  2. Open your account's API keys and copy the key (it ends in :fx).
  3. In Ascent, choose DeepL and paste it. No model and no style: DeepL translates directly.

500,000 characters a month. DeepL is a translator, not a chat model: it does not see the whole page at once.

11. Cerebras

  1. Sign up at cloud.cerebras.ai and add a card to receive the trial credit.
  2. Open API Keys, create one and copy it.
  3. In Ascent, choose Cerebras and paste it. Base URL https://api.cerebras.ai/v1, model gpt-oss-120b.

$5 of credit, which expires 30 days after it is given. Cerebras's page is unclear on whether the trial works without the card.

12. DeepSeek

  1. Sign up at platform.deepseek.com and top up a small amount.
  2. Open API keys, create one and copy it.
  3. In Ascent, choose DeepSeek and paste it. Base URL https://api.deepseek.com/v1, model deepseek-flash.

No free tier, but among the cheapest: about $0.15 per million tokens in and $0.60 out, double on weekday peak hours (01:00–04:00 and 06:00–10:00 UTC).

13. Alibaba Qwen (Model Studio)

  1. Sign in to Alibaba Cloud's international Model Studio (Singapore), complete the account details and accept the terms.
  2. Open API Key, tap Create API key and copy it.
  3. In Ascent, choose Alibaba (Qwen) and paste it. Base URL https://dashscope-intl.aliyuncs.com/compatible-mode/v1, model qwen3.8-flash. If Model Studio shows a different base URL for your workspace, use that one.

New accounts get 1 million free tokens a model for 90 days, but Alibaba has removed this from many models: the console says which still have it. Turning on Free Quota Only stops it charging you. Keys from the Singapore, US and China regions do not work in each other's addresses.

14. SiliconFlow

  1. Sign up at cloud.siliconflow.com.
  2. Open API Keys, create one and copy it.
  3. In Ascent, choose SiliconFlow and paste it. Base URL https://api.siliconflow.com/v1, model Qwen/Qwen3-8B.

SiliconFlow's model list, with which models are free, is shown only after signing in; we could not confirm it.

15. Together AI

  1. Sign up at api.together.ai and buy at least $5 of credit.
  2. Open Settings → API Keys and copy the key.
  3. In Ascent, choose Together AI and paste it. Base URL https://api.together.xyz/v1, model Qwen/Qwen3.8-Flash.

16. Fireworks AI

  1. Sign up at fireworks.ai and add credit.
  2. Open your account's API Keys, create one and copy it.
  3. In Ascent, choose Fireworks AI and paste it. Base URL https://api.fireworks.ai/inference/v1, model accounts/fireworks/models/gpt-oss-120b.

17. Hugging Face

  1. Sign in at huggingface.co.
  2. Open a new fine-grained token, tick Make calls to Inference Providers, create it and copy it.
  3. In Ascent, choose Hugging Face and paste it. Base URL https://router.huggingface.co/v1, model openai/gpt-oss-120b:cheapest, which sends it to the cheapest provider of that model.

Free accounts no longer get monthly credit; PRO includes $2 a month, and beyond that you pay the providers' own prices.

18. OpenAI (ChatGPT)

  1. Sign in at platform.openai.com and add prepaid credit under Billing. A ChatGPT subscription does not include the API.
  2. Open API keys, create a secret key and copy it.
  3. In Ascent, choose OpenAI (ChatGPT) and paste it. The model is gpt-4o-mini, which OpenAI retires on 14 January 2027; gpt-6-luna is the one it names to replace it.

19. Anthropic (Claude)

  1. Sign up at platform.claude.com and verify a phone number to claim the $5 trial credit, or add billing.
  2. Open API keys, create one and copy it.
  3. In Ascent, choose Anthropic (Claude) and paste it. The model is claude-sonnet-5; type claude-haiku-5-5 for a cheaper, faster one.

20. xAI (Grok)

  1. Sign up at console.x.ai and add credit under billing.
  2. Open API keys, create one and copy it.
  3. In Ascent, choose xAI (Grok) and paste it. Base URL https://api.x.ai/v1, model grok-4.3.

21. Moonshot (Kimi)

  1. Sign up at platform.kimi.ai and top up at least $1.
  2. Open API Keys in the console, create one and copy it.
  3. In Ascent, choose Moonshot (Kimi) and paste it. Base URL https://api.moonshot.ai/v1, model kimi-k3.

The Kimi K2 models are closed to accounts made after 31 October 2026.

Your own server, or another service

Any service that speaks OpenAI's chat protocol works: choose OpenAI-compatible at the bottom of the list and type its base URL (ending in /v1, usually) and a model name.

Ollama (http://your-computer:11434/v1, with OLLAMA_HOST=0.0.0.0 set so other devices can reach it) and LM Studio (http://your-computer:1234/v1, with Serve on Local Network on) run models on your own computer for free. Neither serves https://, which Ascent requires because your key and your text travel on the address: put the server behind a tunnel or proxy that gives it an https:// address (Cloudflare Tunnel, ngrok, or Nginx with a certificate). They take no key; type any word as the key. A model of the Qwen family is a good start for Korean, Japanese and Chinese.

Not supported: Azure OpenAI, which takes its key in another way, and GitHub Models, which GitHub closed on 30 July 2026.

If it stops working