The core value is typically 1,000,000 tokens per available model for 90 days (Singapore region), plus 100 images for qwen-image-max and qwen-image-edit-max, and 50 seconds of video for the wan2.6 video models. Because the token quota is granted per model (and not shared), you can test several Qwen variants without them cannibalizing each other. In practical terms, that’s enough for a real prototype: chat flows, toolcalling experiments, and a decent amount of multimodal testing. The “worth” depends on which models you lean on and how output-heavy your prompts are, but it’s far more than a tiny demo tier.

No.

The free quota is typically valid for 90 days from activation for activations after September 8, 2025, and the clock starts the moment you activate Model Studio.

Yes. If you have Alibaba Cloud credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Alibaba Cloud credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

Once your free quota expires or is depleted, usage can be billed at standard rates unless you enable “Free Quota Only,” which forces requests to error instead of charging you.

Singapore (International Edition) is the only region that includes the free token quota. Mainland China (Beijing) does not include it, and the US (Virginia) region is typically available without the free quota.

Yes, it’s compatible with the OpenAI SDK. You swap in Alibaba’s base URL for Singapore (https://dashscope-intl.aliyuncs.com/compatible-mode/v1) and use your DashScope API key as the API key. There’s also a newer Responses API with a different base URL (https://dashscope-intl.aliyuncs.com/api/v2/apps/protocols/compatible-mode/v1), so double-check which API style your code is using. If you already have an OpenAI-based integration, this can be a fast way to evaluate Qwen models without rewriting your whole client.

Students can receive up to $100 in promotional credits (often lower if your institution isn’t a member), and educators can receive up to $200. The credits are meant for learning and coursework inside an AWS Educate Starter Account, alongside the included labs and training.

No.

AWS Educate promotional credits are valid for up to 12 months.

Yes. If you have AWS credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted AWS credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

In AWS Educate, your credits simply stop being usable after their validity window (up to 12 months). Because the Starter Account cannot be billed, you won’t get surprise charges when credits are gone. What you will hit is access friction: if your Starter Account expires after graduation or you don’t renew, you may need to move to a regular AWS account to keep building. For long-term projects, plan that transition early and don’t wait until the last week.

It depends on whether your institution is an AWS Educate member: students at member institutions can get $100, while non-members typically get about $25–$35.

The Starter Account is a restricted learning sandbox: no credit card, us-east-1 only, about 75% of services, and no IAM or billing dashboard. EC2 is also limited (including instance types and about 20 concurrent instances), and sessions time out after about 3 hours of activity. A regular AWS account is the opposite: full regions and services, full IAM and billing, but you can be charged if you exceed free-tier limits. That’s why Educate is safer for beginners, but less useful for production work.

They’re worth 1,000,000 inference tokens per day at $0, which is plenty for serious prototyping and early-stage pilots as long as you stay within the 8,192-token context limit.

No.

The free tier resets daily: you get 1M tokens per day, unused tokens don’t roll over, and you start fresh after the daily reset.

Yes. If you have Cerebras credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Cerebras credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

On the free tier, you don’t “expire” so much as you run out for the day: once you hit the daily cap, requests are restricted until the next daily reset (or until you upgrade).

Mostly, yes: you point your OpenAI SDK to https://api.cerebras.ai/v1 and swap in your Cerebras API key. Cerebras supports the OpenAI-compatible Chat Completions endpoint, not the Text Completions endpoint. A few parameters aren’t supported (frequency_penalty, presence_penalty, logit_bias). Also note the streaming + JSON mode limitation on reasoning models, while streaming works fine with gpt-oss-120b and non-reasoning models.

Production models include Llama 3.1 8B (llama3.1-8b) and OpenAI GPT-OSS 120B (gpt-oss-120b), plus others listed in the Cerebras console. Yes, some are being phased out: Llama 3.3 70B and Qwen 3 32B are scheduled for deprecation on February 16, 2026.

You get 10,000 Neurons per day, which Cloudflare estimates is roughly 1,300 LLM responses, about 12,500 embeddings, around 2,000 small image generations (512×512), or about 8,300 image classifications. Neuron cost varies by model, so bigger models will burn through the daily budget much faster. The good news is the allowance resets every day at 00:00 UTC, so you can treat it like a daily “compute stipend.” If you’re building a demo, split workloads: use smaller models for most requests and reserve the heavy one for the handful of cases that actually need it.

No.

There’s no time limit, but the 10,000 Neurons reset daily at 00:00 UTC.

Yes. If you have Cloudflare credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Cloudflare credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

For Workers AI’s free tier, they don’t “expire”; the daily neuron quota simply resets at 00:00 UTC.

Yes, Workers AI supports OpenAI-compatible endpoints for chat completions, text completions, and embeddings, so you can swap the base URL and keep standard parameters like temperature and max_tokens.

Yes. Cloudflare notes that wrangler dev still consumes your Cloudflare account’s neuron quota, so local testing can quietly burn your daily allowance.

They’re worth 1,000 API requests per calendar month across Cohere endpoints (Chat, Embed, Rerank, and others). Since each request counts as one call regardless of token count, the “value” depends on how chatty your integration is. Practically, it’s enough to build and demo a small prototype, run embedding and rerank experiments, and do prompt iteration without paying. It’s not enough for load testing or anything resembling production traffic.

No.

The 1,000-call cap resets every calendar month, and the trial doesn’t expire as long as your Cohere account exists.

Yes. If you have Cohere credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Cohere credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

When you hit the monthly cap, the Trial key returns errors until the next calendar month reset; there’s no auto-billing because you haven’t added payment info.

No. The Trial key gives access to all Cohere models and endpoints, identical to a Production key.

No, Cohere explicitly prohibits production or commercial use on the Trial key. You can use it for prototyping, learning, research, and proof-of-concept work, but you can’t serve it to end users in a revenue-generating application. If you need to go live, upgrade to a Production key in the dashboard and add payment information. For sensitive use cases (healthcare, legal, employment decisions, and similar), Cohere notes your Production key may remain rate-limited until a manual safety review is completed.

They’re worth about $8.40 in API usage, which you can spend on deepseek-chat or deepseek-reasoner. Because DeepSeek pricing is very low, that can cover a surprising amount of prototyping, especially if your prompts benefit from cache-hit pricing and you keep output tokens under control.

No.

They last 30 days, starting at registration (not your first API call).

Yes. If you have DeepSeek credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted DeepSeek credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

There’s no auto-billing; once the free credits expire (or you use them up), API calls stop working until you top up.

The API currently serves two endpoints powered by DeepSeek-V3.2: deepseek-chat and deepseek-reasoner, and the free tokens apply to API usage.

Yes, mostly. DeepSeek is designed as a drop-in replacement for OpenAI’s API: you set the base URL to https://api.deepseek.com (or /v1 for full compatibility), authenticate with a Bearer API key, and call POST /chat/completions. If you already use the official OpenAI Python or Node.js SDKs, the main changes are the base URL and the model name. That also means many existing wrappers work without major rewrites, which is handy when you’re just trying to evaluate cost and response quality quickly.

$1 in usage on Fireworks’ inference APIs. In practical terms, that can be about 10M tokens on small (<4B) models, roughly 700 FLUX.1 [schnell] images, around 18 hours of Whisper V3 Turbo transcription, or a very large embeddings run (roughly 125M tokens on the cheapest tier). Fireworks also publishes example costs for specific models like DeepSeek V3 and Qwen3 235B, so you can sanity-check your own test plan against their token pricing.

No.

They last until you use the $1, and if you don’t add a payment method your account is suspended once it’s exhausted.

Yes. If you have Fireworks AI credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Fireworks AI credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

If you don’t have a payment method, your account is suspended when the $1 is exhausted, and you won’t be charged.

Without a payment method you’re limited to 10 requests per minute. Add a card and you can go up to about 600 RPM (with a 6,000 RPM spike arrest ceiling), which makes normal testing and benchmarking much less painful.

Yes. Fireworks uses an OpenAI-compatible API format: you typically just change the base URL to https://api.fireworks.ai/inference/v1, generate an API key, and select a Fireworks model identifier. It supports common tooling like OpenAI Python/Node plus popular orchestration layers (LangChain, LlamaIndex, Vercel AI SDK, LiteLLM, and others). Honestly, this is the biggest reason the $1 is useful: you can compare providers without rewriting your whole stack.

There isn’t a fixed dollar credit here; the “value” is free, rate-limited inference, like about 50 High-tier requests/day (for example GPT-4o) or about 150 Low-tier requests/day (for example Phi-4). Practically, that’s plenty for prompt iteration, basic evaluations, and a small prototype that only a few people touch. Once you start running automated jobs or team-wide traffic through it, you will hit 429s fast. At that point, switch to pay-as-you-go or BYOK.

No.

The free tier is ongoing, but usage is limited by per-minute and per-day request windows (and a per-request token cap). When you hit the limit, you get HTTP 429 responses until the window resets.

Yes. If you have GitHub credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted GitHub credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

For GitHub Models’ free tier specifically, you don’t get charged; you just get HTTP 429 until limits reset, or you upgrade to pay-as-you-go/BYOK for higher throughput.

You need a fine-grained personal access token with the models:read scope (required since May 2025).

No. Calls from GitHub Actions count against your account’s rate limits, so a busy workflow can eat the same daily allowance you planned to use for manual testing.

They’re worth roughly 15–30 T4 GPU hours per week (dynamic) plus up to 12-hour sessions with about 12–13 GB RAM and 16 GB VRAM when a GPU is available. In practice, that’s enough for quantized 7B LLM inference, LoRA/QLoRA fine-tuning on smaller models, Stable Diffusion image generation, and Whisper transcription. The exact “value” changes with availability because GPU access is not guaranteed at peak times. If you treat it like a disposable GPU sandbox for experiments and demos, it’s one of the better free deals out there.

No. Colab’s free tier does not require a credit card or any payment method.

A Colab free-tier session can run up to about 12 hours, and GPU availability is governed by a dynamic weekly quota rather than a fixed published limit.

Yes. If you have Google credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Google credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

When your free GPU quota is exhausted, Colab typically restricts you to CPU-only runtimes, and if a session ends or disconnects, the VM is reclaimed and local files are wiped.

Because the idle timeout is based on interaction with the browser tab (clicking, typing, scrolling), not on whether code is running. If you’re inactive for about 90 minutes, Colab disconnects and reclaims the VM. Plan for it. Save checkpoints to Google Drive, and don’t assume a long training job will finish unattended on free tier.

No. The VM disk is ephemeral, so you need to mount Google Drive or download files to persist them.

There isn’t a fixed dollar credit amount. The value is “ongoing free usage” capped by model-specific rate limits (roughly ~30 RPM, ~6K–12K TPM, and daily token/request caps), which can easily cover a prototype, demo, or internal tool if you design around throttling.

No.

They don’t expire because the free tier is ongoing (not a trial).

Yes. If you have Groq credits you won’t use before they expire, you can list them on AI Credit Mart and sell them at up to 70% of face value. Companies regularly list surplus credits from startup programs and enterprise agreements.

AI Credit Mart has discounted Groq credits available from companies with surplus allocations. Prices are typically 30-70% below retail.

For the free tier specifically, there’s no expiration date; when you hit limits you get throttled (429 errors) until limits reset.

Groq states the free tier gives access to all supported models, including Llama 3.1/3.3, Llama 4 Scout/Maverick (preview), GPT-OSS, Qwen3, Kimi K2 (preview), Whisper, and the agentic groq/compound systems. The exact lineup changes, and preview models may be discontinued without notice, so it’s smart to check your console for the current list before you commit to one model in your code. If you care about stability, prefer the “Production Models” section over preview. Also note that different models have different RPM/TPM/TPD caps.

No; exceeding limits returns a 429 Too Many Requests error, and you’re never charged unless you explicitly add payment info and upgrade to the Developer tier.