how to read this
I add to this page as I go, newest entry first. Many entries are a snapshot of prices or model names, and those go out of date. The date on each entry matters more than the date at the top.
I match things by job and price. I haven’t run benchmarks of my own unless an entry says so. Nothing here is advice, so check the sources before you rely on a number.
- API credit I’ve loaded, Jan–Oct 2026 · checked 18 Oct 2026
- Open models on Ollama’s cloud, matched to Claude · checked 18 Oct 2026
api credit i’ve loaded
Checked
These are the pay-as-you-go API services I put prepaid credit into between January and October 2026, leaving out Claude. I wrote them down so I can find them again and see which ones I still use.
There are no amounts on purpose. What each one is for says more than what I paid.
| Service | Type | What I used it for |
|---|---|---|
| Language models · backup and testing | ||
| OpenAI API | Prepaid credit balance | Backup model for when Claude is down, including the direct route to GPT-5.6 Sol. |
| OpenRouter | Credits for 400+ models | One key for my backup models (GPT-6.1 Sol, DeepSeek V4 Pro) and a second route to Claude if Anthropic’s own servers are down. |
| Moonshot AI (Kimi) | Account top-up | Tried Kimi K3 directly and through OpenRouter. It is not in my backup list now. |
| xAI (Grok / SpaceXAI) | Prepaid tokens | A test account. Barely used since. |
| Together AI | Prepaid commit | A test account for open models. The credit ran out within a day. |
| Fireworks AI | API usage credit | A test account for open models. |
| Voice, music and video | ||
| ElevenLabs | Prepaid credits (voice) | The “Alice” voice for a training explainer, and a song test against other music tools. |
| Mureka | API (music) | Songs from my own lyrics in English and Mandarin. Kept as a backup to my main song tool. |
| HeyGen | API (avatar video) | Speech and avatar video tests. I use other tools for narration. |
| Fish Audio (Hanabi AI) | API credit (voice) | Voiceovers and Singapore voices. I compared three models by ear and picked one for finished work. |
From my own receipts and account emails, as of 18 Oct 2026.
What this does not tell you
- Loading credit is not a verdict. I have barely used some of these.
- Some of these cost more than they look. Several add Singapore GST on top of the credit, and unused credit may expire. Check each provider’s terms.
- Your text goes to each provider’s servers. I keep private or work material on a model that runs on my own computer.
- Names and prices change. This list will go out of date.
Setting up API access for your team? Message me on LinkedIn.
open models matched to claude
Checked
When Claude is down I want a backup that doesn’t depend on Anthropic. Ollama’s cloud lists 20 open models. I sorted them by the job they do and what they cost, and put the Claude model of the same tier next to each group.
This is a match on role and price. It is not proof that any of them is as good as Claude.
| Model on Ollama’s cloud | Per million tokens, in / out |
|---|---|
| Hard work · Claude Opus 5.5 is $4 / $20 | |
| Kimi K3Ollama calls it its most capable listing. Built for agent work and takes images. | $3.00 / $15.00 |
| DeepSeek V4 ProA large general model with three reasoning modes. | $1.32 / $3.96 |
| GLM-5.3Z.ai’s flagship, aimed at coding and long multi-step tasks. | $1.40 / $4.40 |
| Muse Spark OpenA very large model that takes up to a million tokens at once. Built for long tasks. | $1.25 / $4.25 |
| Everyday work · Claude Sonnet 5.5 is $2 / $10 | |
| GLM-5.3-flashThe smaller GLM. Its maker says it comes close to an older Claude Opus on coding tests. | $0.15 / $0.50 |
| MiniMax M3Coding and agent work, with a million-token input. | $0.60 / $2.40 |
| Mistral Large 4A general-purpose model that also reads images. | $0.68 / $2.09 |
| DeepSeek V4.1 FlashA faster, cheaper DeepSeek. | $0.30 / $1.20 |
| Quick jobs · Claude Haiku 5.5 is $0.10 / $0.50 | |
| Gemma 4Google’s open model. The same one can run on a laptop. | $0.14 / $0.40 |
| gpt-ossOpenAI’s open model. | from $0.07 / $0.30 |
| Nemotron 3 NanoNVIDIA’s small model. | $0.06 / $0.24 |
Ollama prices are from its cloud list. Claude prices are OpenRouter’s list prices. Both as of 18 Oct 2026, and both change.
What this does not tell you
- Price is not quality. A cheaper model can be weaker on your work.
- “Close to Claude” is the maker’s claim. It compares with an older Claude. I haven’t checked it.
- Most of these need a paid plan. When I tried the free Ollama plan, only the cloud Gemma 4 ran. The others asked for the Pro plan or bought credits.
- Cloud means your text goes to servers. Ollama says it hosts the models mostly in the United States, uses partner data centres run by NVIDIA cloud providers, and does not log or train on prompts. That is Ollama’s own statement. I keep private or work material on a model that runs on my own computer.
- You may not need Ollama’s cloud at all. Kimi K3, DeepSeek V4 Pro and GLM-5.3 are also sold through other services such as OpenRouter.
A fair test
Give three of them the same real task you would give Claude and read the answers side by side. Judge on your own work, not on a leaderboard. I’ll add my results here when I run it.
Wondering how you’d set up a backup for your team? Message me on LinkedIn.