Models and prices.
Every model runs on servers in the EU. Prices in euros per million tokens — you pay for the model that actually answered.
Start here
Three names for agents. They point at the best model for the job today, so you never change your configuration when we upgrade.
soloniq/agentReasoning and tool calls — the default for agents.
€0.20€0.238 · €0.05€0.0595 · €0.50€0.595
input · cached input · output, per million tokens
Runs on GLM-5.3-Flash; fallback: DeepSeek V4.1 Flash
soloniq/agent-fastLight reasoning: summaries, routine steps, context compression.
€0.20€0.238 · €0.05€0.0595 · €0.50€0.595
input · cached input · output, per million tokens
Runs on GLM-5.3-Flash (low reasoning); fallback: DeepSeek V4.1 Flash
soloniq/agent-proThe stronger tier: harder tasks and long agent runs.
€1.75€2.0825 · €0.44€0.5236 · €4.50€5.355
input · cached input · output, per million tokens
Runs on GLM-5.3; fallback: DeepSeek V4 Pro
All models
| Model | Context | Images | Input | Cached input | Output | 1,000 agent turns* | Fallback |
|---|---|---|---|---|---|---|---|
GLM-5.3-Flash Z.aisoloniq/glm-5.3-flashFast and capable; the default behind agent and agent-fast. |
1M | — | €0.20€0.238 | €0.05€0.0595 | €0.50€0.595 | €1.55€1.84 | deepseek-v4.1-flash |
Qwen 3.8 Flash Alibaba Qwensoloniq/qwen-3.8-flashFast and cheap, sees images. |
256K | yes | €0.20€0.238 | €0.05€0.0595 | €0.50€0.595 | €1.55€1.84 | glm-5.3-flash |
DeepSeek V4 Flash DeepSeeksoloniq/deepseek-v4-flashLowest output price for high-volume text work. |
1M | — | €0.25€0.2975 | €0.06€0.0714 | €0.30€0.357 | €1.73€2.06 | deepseek-v4.1-flash |
MiniMax M3 MiniMaxsoloniq/minimax-m3Long-context agents, sees images. |
1M | yes | €0.40€0.476 | €0.10€0.119 | €2.00€2.38 | €3.60€4.28 | deepseek-v4.1-flash |
DeepSeek V4.1 Flash DeepSeeksoloniq/deepseek-v4.1-flashStrong reasoning, sees images; the agent backup. |
1M | yes | €0.50€0.595 | €0.13€0.1547 | €1.50€1.785 | €4.09€4.87 | — |
Kimi K2.7 Code Moonshot AIsoloniq/kimi-k2.7-codeCoding specialist. |
256K | yes | €1.25€1.4875 | €0.31€0.3689 | €4.50€5.355 | €10.33€12.29 | glm-5.3 |
GLM-5.3 Z.aisoloniq/glm-5.3Z.ai flagship for coding and long agent runs. |
1M | — | €1.75€2.0825 | €0.44€0.5236 | €4.50€5.355 | €13.67€16.27 | deepseek-v4-pro |
DeepSeek V4 Pro DeepSeeksoloniq/deepseek-v4-proHard reasoning, maths and code. |
1M | — | €2.00€2.38 | €0.50€0.595 | €4.00€4.76 | €15.00€17.85 | glm-5.3 |
Qwen 3.8 Alibaba Qwensoloniq/qwen-3.8Qwen flagship for demanding reasoning. |
256K | — | €2.50€2.975 | €0.63€0.7497 | €6.00€7.14 | €19.34€23.01 | glm-5.3 |
Kimi K3 Moonshot AIsoloniq/kimi-k3Premium agent model with vision. |
1M | yes | €3.00€3.57 | €0.75€0.8925 | €15.00€17.85 | €27.00€32.13 | glm-5.3 |
*A typical agent turn: 20,000 input tokens, of which 18,000 are cached, and 500 output tokens.
How billing works
- Per token, per model. Input, cached input and output tokens are counted separately and billed at the price of the model that answered — shown in the
modelfield of every reply. - Cached input is billed at the cached price whenever the model reports a cache hit (
usage.prompt_tokens_details.cached_tokens). Agents resend their context constantly, so this is usually most of the bill. - Thinking counts as output. Reasoning tokens are part of
completion_tokens. - Fallback. If a model fails or is slow, its fallback answers and is billed at its own price.
- Images go to an image-capable model in the chain; if there is none, the request is refused with
vision_not_supported. - Prepaid in euros. Direct customers top up from €30 by PayPal (plus a 2.5% PayPal fee) or bank transfer, with an invoice for every payment; partners receive one monthly invoice. Business prices exclude VAT; private customers pay prices including 19% VAT.