Radox AI Models 2026: The 33-Model Catalog, Rates per 1M Tokens and How to Choose
Radox AI lists 33 chat models behind one OpenAI-compatible endpoint, from 10 makers, and every plan reaches all of them. This guide turns the official models page into a readable catalog: model ID, context window, listed rate per 1M tokens and capability labels, followed by a short way to pick one. This article focuses on chat/text.
| Models | 33, one endpoint |
| Makers | 10 |
| Listed input rate | $0.115 to $16.606 per 1M |
| Context window | 8K to 1050K |
| Vision input | 28 models |
| Built-in web search | 11 models |
A long catalog is only useful if you can shrink it quickly.
Three columns do most of the work: context window, listed rate and the capability labels. Everything else is a name.
The official models page opens with a plain promise: "Every model below speaks the same OpenAI-compatible API." and "Switch by changing one string". That makes choosing a model a low-stakes decision, because you can change your mind by editing one value in your code. The useful question is therefore not "which model is best" but "which two or three do I try first".
33 models, 10 makers, and a rate spread of about 144x.
All typed "chat" in the catalog.
Anthropic has the most entries (9).
Per 1M tokens, across the catalog.
Largest listed value in the catalog.
The spread is calculated as the highest listed input rate ($16.606) divided by the lowest ($0.115), rounded.
Ten makers, one list: who is in the catalog.
| Maker | Models | IDs | Listed input rate per 1M |
|---|---|---|---|
| Anthropic | 9 | claude-sonnet-5, claude-mythos-preview, claude-opus-4.6, claude-opus-4.7, claude-opus-4.8, claude-opus-5, claude-opus-5.5, claude-fable-5, claude-fable-5.1 | $3.324 to $16.606 |
| OpenAI | 5 | gpt-5.6-luna, gpt-5.6-terra, gpt-5.5, gpt-5.5-pro, gpt-5.6-sol | $0.357 to $8.947 |
| 4 | gemma-2-2b, gemini-3.7-flash, gemini-3.6-flash, gemini-3.1-pro | $0.115 to $4.474 | |
| DeepSeek | 3 | deepseek-v4-flash-0731, deepseek-v4-pro-0813, deepseek-v4-pro-max | $0.115 to $2.047 |
| Alibaba Qwen | 3 | qwen3.8-27b, qwen3.7-max, qwen3.8-max | $0.575 to $1.76 |
| ByteDance Seed | 2 | seed-2.1-turbo, seed-2.1-pro | $0.92 to $1.15 |
| Moonshot AI | 2 | kimi-k2.6, kimi-k3 | $1.219 to $4.98 |
| Zhipu AI | 2 | glm-5.2, glm-5.3 | $1.357 |
| xAI | 2 | grok-4.5, grok-4.6 | $2.806 |
| Meta | 1 | muse-spark-1.1 | $1.817 |
Swipe sideways to see every column.
Counting labels per maker shows where each capability sits, and the averages and spreads show how differently the makers price their own lineups.
| Maker | Models | Reasoning | Vision | Tools | Web |
|---|---|---|---|---|---|
| Anthropic | 9 | 9 | 9 | 9 | 0 |
| OpenAI | 5 | 5 | 5 | 5 | 0 |
| 4 | 3 | 3 | 3 | 3 | |
| DeepSeek | 3 | 3 | 0 | 3 | 0 |
| Alibaba Qwen | 3 | 3 | 2 | 3 | 2 |
| ByteDance Seed | 2 | 2 | 2 | 2 | 0 |
| Moonshot AI | 2 | 2 | 2 | 2 | 2 |
| Zhipu AI | 2 | 2 | 2 | 2 | 2 |
| xAI | 2 | 2 | 2 | 2 | 2 |
| Meta | 1 | 1 | 1 | 1 | 0 |
| All makers | 33 | 32 | 28 | 32 | 11 |
All 33 models, sorted by listed rate.
This is the whole catalog as listed on the models page. Rates are Radox's listed price per 1M tokens, rounded to three decimals the way the homepage shows them. The catalog's own capability labels are reproduced as-is.
Swipe sideways to see every column.
| Model ID | Maker | Context | In / 1M | Out / 1M | Listed capabilities |
|---|---|---|---|---|---|
deepseek-v4-flash-0731 | DeepSeek | 1049K | $0.115 | $0.115 | Reasoning, Tools, Streaming, JSON |
gemma-2-2b | 8K | $0.115 | $0.23 | Streaming, JSON | |
gpt-5.6-luna | OpenAI | 1050K | $0.357 | $0.357 | Reasoning, Vision, Tools, Streaming, JSON |
deepseek-v4-pro-0813 | DeepSeek | 1049K | $0.552 | $0.552 | Reasoning, Tools, Streaming, JSON |
qwen3.8-27b | Alibaba Qwen | 262K | $0.575 | $0.575 | Reasoning, Tools, Streaming, JSON |
seed-2.1-turbo | ByteDance Seed | 262K | $0.92 | $0.92 | Reasoning, Vision, Tools, Streaming, JSON |
seed-2.1-pro | ByteDance Seed | 262K | $1.15 | $1.15 | Reasoning, Vision, Tools, Streaming, JSON |
kimi-k2.6 | Moonshot AI | 1000K | $1.219 | $1.219 | Reasoning, Vision, Tools, Streaming, JSON, Web |
gemini-3.7-flash | 1049K | $1.242 | $1.242 | Reasoning, Vision, Tools, Streaming, JSON, Web | |
glm-5.2 | Zhipu AI | 1000K | $1.357 | $1.357 | Reasoning, Vision, Tools, Streaming, JSON, Web |
glm-5.3 | Zhipu AI | 1000K | $1.357 | $1.357 | Reasoning, Vision, Tools, Streaming, JSON, Web |
qwen3.7-max | Alibaba Qwen | 1000K | $1.76 | $1.76 | Reasoning, Vision, Tools, Streaming, JSON, Web |
qwen3.8-max | Alibaba Qwen | 1000K | $1.76 | $1.76 | Reasoning, Vision, Tools, Streaming, JSON, Web |
muse-spark-1.1 | Meta | 1000K | $1.817 | $1.817 | Reasoning, Vision, Tools, Streaming, JSON |
deepseek-v4-pro-max | DeepSeek | 1049K | $2.047 | $2.047 | Reasoning, Tools, Streaming, JSON |
gemini-3.6-flash | 1049K | $2.496 | $2.496 | Reasoning, Vision, Tools, Streaming, JSON, Web | |
grok-4.5 | xAI | 500K | $2.806 | $2.806 | Reasoning, Vision, Tools, Streaming, JSON, Web |
grok-4.6 | xAI | 500K | $2.806 | $2.806 | Reasoning, Vision, Tools, Streaming, JSON, Web |
claude-sonnet-5 | Anthropic | 1000K | $3.324 | $3.324 | Reasoning, Vision, Tools, Streaming, JSON |
gpt-5.6-terra | OpenAI | 1050K | $3.577 | $3.577 | Reasoning, Vision, Tools, Streaming, JSON |
gemini-3.1-pro | 1049K | $4.474 | $4.474 | Reasoning, Vision, Tools, Streaming, JSON, Web | |
kimi-k3 | Moonshot AI | 1000K | $4.98 | $4.98 | Reasoning, Vision, Tools, Streaming, JSON, Web |
claude-mythos-preview | Anthropic | 1000K | $8.303 | $8.303 | Reasoning, Vision, Tools, Streaming, JSON |
claude-opus-4.6 | Anthropic | 1000K | $8.303 | $8.303 | Reasoning, Vision, Tools, Streaming, JSON |
claude-opus-4.7 | Anthropic | 1000K | $8.303 | $8.303 | Reasoning, Vision, Tools, Streaming, JSON |
claude-opus-4.8 | Anthropic | 1000K | $8.303 | $8.303 | Reasoning, Vision, Tools, Streaming, JSON |
claude-opus-5 | Anthropic | 1000K | $8.303 | $8.303 | Reasoning, Vision, Tools, Streaming, JSON |
claude-opus-5.5 | Anthropic | 1000K | $8.303 | $8.303 | Reasoning, Vision, Tools, Streaming, JSON |
gpt-5.5 | OpenAI | 1050K | $8.947 | $8.947 | Reasoning, Vision, Tools, Streaming, JSON |
gpt-5.5-pro | OpenAI | 1050K | $8.947 | $8.947 | Reasoning, Vision, Tools, Streaming, JSON |
gpt-5.6-sol | OpenAI | 1050K | $8.947 | $8.947 | Reasoning, Vision, Tools, Streaming, JSON |
claude-fable-5 | Anthropic | 1000K | $16.606 | $16.606 | Reasoning, Vision, Tools, Streaming, JSON |
claude-fable-5.1 | Anthropic | 1000K | $16.606 | $16.606 | Reasoning, Vision, Tools, Streaming, JSON |
For 32 of the 33 models the listed output rate equals the input rate. The exception is gemma-2-2b: $0.115 in, $0.23 out.
The same 33 rates as bars
Context windows at a glance
Most of the catalog lists a window of 1000K or more. Only six models sit below that.
The models page and every model on one plane
- Anthropic
- OpenAI
- DeepSeek
- Alibaba Qwen
- ByteDance Seed
- Moonshot AI
- Zhipu AI
- xAI
- Meta
Reading table for the crowded 1000K+ row (27 models)
| Model ID | In / 1M | Context |
|---|---|---|
deepseek-v4-flash-0731 | $0.115 | 1049K |
gpt-5.6-luna | $0.357 | 1050K |
deepseek-v4-pro-0813 | $0.552 | 1049K |
kimi-k2.6 | $1.219 | 1000K |
gemini-3.7-flash | $1.242 | 1049K |
glm-5.3 | $1.357 | 1000K |
glm-5.2 | $1.357 | 1000K |
qwen3.7-max | $1.760 | 1000K |
qwen3.8-max | $1.760 | 1000K |
muse-spark-1.1 | $1.817 | 1000K |
deepseek-v4-pro-max | $2.047 | 1049K |
gemini-3.6-flash | $2.496 | 1049K |
claude-sonnet-5 | $3.324 | 1000K |
gpt-5.6-terra | $3.577 | 1050K |
gemini-3.1-pro | $4.474 | 1049K |
kimi-k3 | $4.980 | 1000K |
claude-opus-5.5 | $8.303 | 1000K |
claude-opus-4.7 | $8.303 | 1000K |
claude-opus-4.6 | $8.303 | 1000K |
claude-opus-4.8 | $8.303 | 1000K |
claude-mythos-preview | $8.303 | 1000K |
claude-opus-5 | $8.303 | 1000K |
gpt-5.5-pro | $8.947 | 1050K |
gpt-5.5 | $8.947 | 1050K |
gpt-5.6-sol | $8.947 | 1050K |
claude-fable-5 | $16.606 | 1000K |
claude-fable-5.1 | $16.606 | 1000K |
What the six labels mean, and how many models carry each.
The models page defines each label in a short legend. The count column is calculated from the catalog table above.
| Label | Radox's meaning | Models with it |
|---|---|---|
| Reasoning | Thinks step by step before answering | 32 |
| Vision | Accepts image input | 28 |
| Tools | Function / tool calling | 32 |
| Streaming | Server-sent event streaming | 33 |
| JSON | Structured JSON output | 33 |
| Web | Built-in web search | 11 |
The page adds that "Capabilities are returned by GET /v1/models", so your code can check a label at runtime instead of hard-coding it. A label tells you the model accepts that kind of input or feature; it does not tell you how well it does the job.
Vision and Web together
The two labels people filter on most are Vision and Web. This is how they overlap.
| Labels carried | Models | IDs |
|---|---|---|
| Vision and Web | 11 | kimi-k2.6, gemini-3.7-flash, glm-5.2, glm-5.3, qwen3.7-max, qwen3.8-max, gemini-3.6-flash, grok-4.5, grok-4.6, gemini-3.1-pro, kimi-k3 |
| Vision only | 17 | gpt-5.6-luna, seed-2.1-turbo, seed-2.1-pro, muse-spark-1.1, claude-sonnet-5 and 12 more |
| Web only | 0 | none |
| Neither | 5 | deepseek-v4-flash-0731, gemma-2-2b, deepseek-v4-pro-0813, qwen3.8-27b, deepseek-v4-pro-max |
The 11 models with built-in web search
| Model ID | Maker | Context | In / 1M | Vision |
|---|---|---|---|---|
kimi-k2.6 | Moonshot AI | 1000K | $1.219 | Yes |
gemini-3.7-flash | 1049K | $1.242 | Yes | |
glm-5.2 | Zhipu AI | 1000K | $1.357 | Yes |
glm-5.3 | Zhipu AI | 1000K | $1.357 | Yes |
qwen3.7-max | Alibaba Qwen | 1000K | $1.76 | Yes |
qwen3.8-max | Alibaba Qwen | 1000K | $1.76 | Yes |
gemini-3.6-flash | 1049K | $2.496 | Yes | |
grok-4.5 | xAI | 500K | $2.806 | Yes |
grok-4.6 | xAI | 500K | $2.806 | Yes |
gemini-3.1-pro | 1049K | $4.474 | Yes | |
kimi-k3 | Moonshot AI | 1000K | $4.98 | Yes |
Swipe sideways to see every column.
The models that do not accept images
The five that carry neither Vision nor Web: deepseek-v4-flash-0731, gemma-2-2b, deepseek-v4-pro-0813, qwen3.8-27b and deepseek-v4-pro-max (listed rates $0.115 to $2.047, from the cross-count above).
| Context: models | Reason | Vision | Tools | Web | JSON | Stream |
|---|---|---|---|---|---|---|
| 1000K+: 27 | 27 | 24 | 27 | 9 | 27 | 27 |
| 500K: 2 | 2 | 2 | 2 | 2 | 2 | 2 |
| 262K: 3 | 3 | 2 | 3 | 0 | 3 | 3 |
| 8K: 1 | 0 | 0 | 0 | 0 | 1 | 1 |
Four price bands, and most of the catalog sits in the middle two.
Bands are not tiers Radox sells. They are a quick way to see how the list is spread: 6 models under $1, 12 from $1 to under $3, 13 from $3 to under $9, and 2 at $9 and above.
Models that share a rate
Several entries list exactly the same number, so moving between them changes nothing on the bill.
| Listed rate per 1M | Models sharing it | IDs |
|---|---|---|
| $0.115 | 2 | deepseek-v4-flash-0731, gemma-2-2b |
| $1.357 | 2 | glm-5.2, glm-5.3 |
| $1.76 | 2 | qwen3.7-max, qwen3.8-max |
| $2.806 | 2 | grok-4.5, grok-4.6 |
| $8.303 | 6 | claude-mythos-preview, claude-opus-4.6, claude-opus-4.7, claude-opus-4.8, claude-opus-5, claude-opus-5.5 |
| $8.947 | 3 | gpt-5.5, gpt-5.5-pro, gpt-5.6-sol |
| $16.606 | 2 | claude-fable-5, claude-fable-5.1 |
What the rate means for your plan. Your monthly allowance is spent first; beyond it, requests bill from your balance at the model's per-token rate. Radox's Terms state that rates "include our margin" and that "Rates can change". Treat the table as a snapshot dated Oct 15, 2026.
Makers by rate band
| Maker | Under $1 | $1 to under $3 | $3 to under $9 | $9 and above | Total |
|---|---|---|---|---|---|
| Anthropic | 0 | 0 | 7 | 2 | 9 |
| OpenAI | 1 | 0 | 4 | 0 | 5 |
| 1 | 2 | 1 | 0 | 4 | |
| DeepSeek | 2 | 1 | 0 | 0 | 3 |
| Alibaba Qwen | 1 | 2 | 0 | 0 | 3 |
| ByteDance Seed | 1 | 1 | 0 | 0 | 2 |
| Moonshot AI | 0 | 1 | 1 | 0 | 2 |
| Zhipu AI | 0 | 2 | 0 | 0 | 2 |
| xAI | 0 | 2 | 0 | 0 | 2 |
| Meta | 0 | 1 | 0 | 0 | 1 |
A short path from 33 down to two or three.
Keep only the 28 models labelled Vision.
Keep the 11 labelled Web.
27 models list a 1000K or larger context window; the 6 others range from 8K to 500K.
Start with the 6 models under $1 per 1M tokens and test whether one is good enough for your task.
Try one model from the upper bands, then compare its output with a lower-band model on your own prompts.
The same questions as filters on the catalog
Stacking filters shrinks the list quickly. Here the first four questions are applied to the catalog in order.
Each line applies one more filter to the line above. Counts calculated by JAVIS from radoxai.com/models, Oct 15, 2026.
What each added requirement does to the lowest listed rate
gemma-2-2b, deepseek-v4-flash-0731. That is 1.0x the lowest rate in the whole catalog.deepseek-v4-flash-0731. That is 1.0x the lowest rate in the whole catalog.gpt-5.6-luna. That is 3.1x the lowest rate in the whole catalog.kimi-k2.6. That is 10.6x the lowest rate in the whole catalog.kimi-k2.6. That is 10.6x the lowest rate in the whole catalog.Filters turned into starting lists.
| If you need | Filter on | Start from |
|---|---|---|
| Image input | Vision label | 28 models, for example gpt-5.6-luna, seed-2.1-turbo, seed-2.1-pro, kimi-k2.6 |
| Built-in web search | Web label | kimi-k2.6, gemini-3.7-flash, glm-5.2, glm-5.3, qwen3.7-max, qwen3.8-max, gemini-3.6-flash, grok-4.5, grok-4.6, gemini-3.1-pro, kimi-k3 |
| Lowest listed rates (under $1 per 1M) | Input rate under $1 per 1M | deepseek-v4-flash-0731, gemma-2-2b, gpt-5.6-luna, deepseek-v4-pro-0813, qwen3.8-27b, seed-2.1-turbo |
| Coding and reasoning, per Radox's notes | Catalog description | deepseek-v4-pro-0813: "Excellent reasoning and coding"; claude-sonnet-5: "Great coding and reasoning at a lower cost"; gpt-5.6-terra: "Strong reasoning and coding at a mid price point" |
One catalog note per maker
| Maker | Model ID | Radox's own catalog note | In / 1M |
|---|---|---|---|
| Anthropic | claude-sonnet-5 | “Great coding and reasoning at a lower cost.” | $3.324 |
| OpenAI | gpt-5.6-terra | “Strong reasoning and coding at a mid price point.” | $3.577 |
gemini-3.1-pro | “Strong reasoning and coding with high Code Arena score.” | $4.474 | |
| DeepSeek | deepseek-v4-pro-0813 | “Excellent reasoning and coding, 1.6T params.” | $0.552 |
| Alibaba Qwen | qwen3.8-max | “Strong all-round capability, 2.4T params.” | $1.76 |
| ByteDance Seed | seed-2.1-pro | “Strong reasoning and coding.” | $1.15 |
| Moonshot AI | kimi-k3 | “Strong reasoning and coding, 2.8T params.” | $4.98 |
| Zhipu AI | glm-5.3 | “Strong reasoning and agentic capability, 753B params.” | $1.357 |
| xAI | grok-4.6 | “Improved reasoning and coding over 4.5.” | $2.806 |
| Meta | muse-spark-1.1 | “Fast and capable with strong reasoning scores.” | $1.817 |
Swipe sideways to see every column.
The model ID is the only thing that changes.
Pick an ID from the table and put it in the model field. The rest of the request stays the same, which is what "Switch by changing one string" means in practice.
What one extra million tokens costs at overage, per model.
The plan allowance is covered by your monthly price. These figures show what tokens beyond the allowance cost when billed from balance at the listed rate. The table counts input tokens at each model's listed input rate. For 32 of the 33 models the output rate is the same; gemma-2-2b lists $0.23 per 1M output, so 100M output tokens would cost $23.00, and 100M split 50/50 between input and output would cost $17.25.
| Model | 1M tokens | 10M tokens | 100M tokens |
|---|---|---|---|
gemma-2-2b | $0.115 | $1.15 | $11.50 |
deepseek-v4-flash-0731 | $0.115 | $1.15 | $11.50 |
gpt-5.6-luna | $0.357 | $3.57 | $35.65 |
claude-sonnet-5 | $3.324 | $33.24 | $332.35 |
claude-opus-5.5 | $8.303 | $83.03 | $830.30 |
claude-fable-5.1 | $16.606 | $166.06 | $1,660.60 |
What $10 of balance buys
Model names are a catalog entry, not a promise.
Radox's Terms describe the service as a gateway to third-party providers and say: "Model names on our platform may differ from the upstream provider's naming." They also say "Models may be added, renamed, repriced, deprecated, or withdrawn". One catalog entry, claude-mythos-preview, is described by Radox itself as "not yet generally released".
- Keep the model ID in one config value, not scattered through code.
- Re-read the catalog before a large job.
- Keep a second candidate model ready.
The Privacy Policy states: "Which provider serves a given model is shown on our models page."
Match the need before comparing catalogs.
| If you mostly need... | Look at |
|---|---|
| A direct contract and terms with the model maker | The maker's own API, such as Anthropic, OpenAI or Google |
| Another hosted multi-model gateway | OpenRouter |
| A gateway you host and control yourself | LiteLLM proxy (open source) |
| A first look at Claude before choosing a model | Our Claude review |
How this guide was made.
Links labelled as Radox AI calls to action on this page use JAVIS's referral link; if you sign up through them, JAVIS may receive a referral reward at no extra cost to you.
This guide is built from Radox AI's official pages (models, home, pricing, docs, Terms, Privacy), dated Oct 15, 2026. This article focuses on chat/text.
Start on the free plan; no card required per the pricing page.
Create a free accountRead the related guides below first.
Choose the next articleIf you are choosing between models and gateways.
Questions people ask about the Radox AI model list
How many models does Radox AI offer?
33, all typed as chat models in the catalog, from 10 makers.
Does the free plan reach every model?
Yes. The pricing page says every plan, including Free, has access to all 33 models; plans differ in token allowance and rate limits.
Which models accept images as input?
The 28 with the Vision label: gpt-5.6-luna, seed-2.1-turbo, seed-2.1-pro, kimi-k2.6, gemini-3.7-flash, glm-5.2, glm-5.3, qwen3.7-max, qwen3.8-max, muse-spark-1.1, gemini-3.6-flash, grok-4.5, grok-4.6, claude-sonnet-5, gpt-5.6-terra, gemini-3.1-pro, kimi-k3, claude-mythos-preview, claude-opus-4.6, claude-opus-4.7, claude-opus-4.8, claude-opus-5, claude-opus-5.5, gpt-5.5, gpt-5.5-pro, gpt-5.6-sol, claude-fable-5, claude-fable-5.1.
Which models have built-in web search?
The 11 with the Web label: kimi-k2.6, gemini-3.7-flash, glm-5.2, glm-5.3, qwen3.7-max, qwen3.8-max, gemini-3.6-flash, grok-4.5, grok-4.6, gemini-3.1-pro, kimi-k3.
What is the largest context window listed?
1050K, shown on 5 of the 33 models.
How do I see capabilities from code?
Radox says capabilities are returned by GET /v1/models, so your code can check them at runtime.
Can model names change?
Yes. The Terms say models may be added, renamed, repriced, deprecated or withdrawn, so recheck the catalog before large jobs.
Where this came from, and the date it applies to.
Models, IDs, context windows, rates, capability labels: radoxai.com/models (catalog data and legend), Oct 15, 2026.
Plans reaching all models, billing order, 402: radoxai.com/pricing, Oct 15, 2026.
Endpoint, quickstart: radoxai.com (home) and radoxai.com/docs, Oct 15, 2026.
Model naming, rates, withdrawal: Radox Terms of Service, effective 23 August 2026.
Provider disclosure: Radox Privacy Policy, effective 23 August 2026.
Calculated figures: counts, bands, ranges and cost rows are calculated by JAVIS from the catalog values above.
Images: logo hosted on radoxai.com (branding).
Added charts and tables: every figure in the charts, matrices and filter counts is calculated by JAVIS from the same models page, Oct 15, 2026; the cost table was recomputed row by row from the exact catalog rates.
JAVIS has not run this service hands-on; everything here comes from Radox AI's own pages and catalog.
