Model Selection
Available Model Options in NinjaCat
"NinjaCat" Model Options
The "NinjaCat" model is our recommended default for mid-range tasks, offering the best balance of price and performance for everyday workloads — the right choice when you want strong quality without paying for a premium frontier model.
We plan to improve the NinjaCat model over time, so choosing it means your workloads automatically benefit from future gains.
| Model | Notes |
|---|---|
| NinjaCat (US) | Default for all newly created agents. NinjaCat selects a high-performing model on your behalf and routes requests through US infrastructure. The underlying model is managed by NinjaCat and may be updated as better options become available — you will not need to make any changes. |
| NinjaCat | Same managed-model approach without a specific data-residency constraint. |
How Your Data Is Handled
- Zero data retention — we do not log or store your prompt or generation data.
- No training on your data — your data is never used to train models and is never shared back with the underlying provider.
- Secure by default — you retain full ownership of your data, which is always encrypted in transit and at rest.
- US data residency (US-only version) — inference is processed entirely within the United States.
Tip: If you are not sure which model to use, start here. NinjaCat (US) is the recommended starting point for most agents and can be changed at any time.
Anthropic Model Options
| Model | Release Date | Context Window | Input $ / 1M | Output $ / 1M | Notes |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Sep 2026 | ~1M | ~$10 | ~$50 | Anthropic's most capable model — best for deep reasoning, long-context work, and nuanced writing. Cached input (cache-read) ~$0.25 / 1M. (2x cost) |
| Claude Fable 5 | Jun 2026 | ~1M | ~$10 | ~$50 | High-capability model for demanding reasoning and long-horizon agentic work. Cached input (cache-read) ~$1.00 / 1M. (2x cost) |
| Claude Opus 5.5 | Sep 2026 | ~1M | ~$4 | ~$20 | Anthropic's latest Opus model for complex agentic and coding work. Cached input ~$0.20 / 1M. |
| Claude Opus 5.5 - Thinking | Sep 2026 | ~1M | ~$4 | ~$20 | Opus 5.5 with adaptive thinking pinned to high effort for deeper reasoning; uses more output tokens. Cached input ~$0.20 / 1M. |
| Claude Opus 5 | Jul 2026 | ~1M | ~$5 | ~$25 | Anthropic's most capable model for complex agentic and coding work. Cached input ~$0.50 / 1M. |
| Claude Opus 5 - Thinking | Jul 2026 | ~1M | ~$5 | ~$25 | Opus 5 with adaptive thinking pinned to high effort for deeper reasoning. Cached input ~$0.50 / 1M. |
| Claude Opus 4.8 | Jun 2026 | ~1M | ~$5 | ~$25 | Highly capable Opus-tier model for complex reasoning and agentic coding. Cached input ~$0.50 / 1M. |
| Claude Opus 4.8 - Thinking | Jun 2026 | ~1M | ~$5 | ~$25 | Opus 4.8 with adaptive thinking for the most demanding tasks. Cached input ~$0.50 / 1M. |
| Claude Opus 4.7 | Apr 2026 | ~1M | ~$5 | ~$25 | Balanced performance & cost with medium reasoning. Cached input ~$0.50 / 1M. |
| Claude Opus 4.7 - Thinking | Apr 2026 | ~1M | ~$5 | ~$25 | High reasoning for complex multi-step tasks and coding. Cached input ~$0.50 / 1M. |
| Claude Opus 4.6 | Feb 2026 | ~1M | ~$5 | ~$25 | Highly capable model for complex reasoning and coding; use for less complex tasks where fewer tokens are used. Cached input ~$0.50 / 1M. |
| Claude Opus 4.6 - Thinking | Feb 2026 | ~1M | ~$5 | ~$25 | Opus 4.6 with extended thinking for the most demanding tasks. Cached input ~$0.50 / 1M. |
| Claude Sonnet 5.5 | Recently added | ~1M | ~$2 | ~$10 | Anthropic's best combination of speed and intelligence. Cached input ~$0.20 / 1M. |
| Claude Sonnet 5.5 - Thinking | Recently added | ~1M | ~$2 | ~$10 | Sonnet 5.5 with adaptive thinking for deeper reasoning. Cached input ~$0.20 / 1M. |
| Claude Sonnet 5 | Jul 2026 | ~1M | ~$3 | ~$15 | Previous-generation Sonnet balancing speed and intelligence. Price updated 2026-09-01. Cached input ~$0.30 / 1M. |
| Claude Sonnet 5 - Thinking | Jul 2026 | ~1M | ~$3 | ~$15 | Sonnet 5 with adaptive thinking for deeper reasoning. Cached input ~$0.30 / 1M. |
| Claude Sonnet 4.6 | Feb 2026 | ~1M | ~$3 | ~$15 | High-performance model balancing speed and intelligence. Cached input ~$0.30 / 1M. |
| Claude Sonnet 4.6 Thinking | Feb 2026 | ~1M | ~$3 | ~$15 | Sonnet 4.6 with extended thinking for deeper reasoning. Cached input ~$0.30 / 1M. |
| Claude Haiku 4.5 | Oct 2025 | ~200K† | ~$1 | ~$5 | Fast and cost-effective model for routine tasks. Cached input ~$0.10 / 1M. |
| Claude Haiku 4.5 Thinking | Oct 2025 | ~200K† | ~$1 | ~$5 | Haiku 4.5 with extended thinking for improved reasoning at low cost. Cached input ~$0.10 / 1M. |
OpenAI Model Options
| Model | Release Date | Context Window | Input $ / 1M | Output $ / 1M | Notes |
|---|---|---|---|---|---|
| GPT-6 Astra | Sep 2026 | 272K | ~$10 | ~$50 | OpenAI's frontier flagship — the top-end choice for the hardest reasoning tasks; higher cost, so reserve it for high-value work. Cached input ~$1.00 / 1M. Long-context pricing above 272K input: ~$20 in / ~$75 out / ~$2.00 cached. (2x cost) |
| GPT-6.1 Sol | Sep 2026 | 272K | ~$2 | ~$10 | Near-Astra performance at a lower cost for reasoning and agentic work. Cached input ~$0.10 / 1M. Long-context pricing above 272K input: ~$4 in / ~$15 out / ~$0.20 cached. |
| GPT-6 Luna | Sep 2026 | 272K | ~$0.10 | ~$0.50 | Fast, low-cost GPT-6 model for high-volume tasks. Cached input ~$0.01 / 1M. Long-context pricing above 272K input: ~$0.20 in / ~$0.75 out / ~$0.02 cached. |
| GPT-5.6 Sol | Aug 2026 | ~1M† | ~$5.00 | ~$30 | Most capable GPT-5.6 tier; built for demanding coding and analysis tasks. Cached input ~$0.50 / 1M. |
| GPT-5.6 Terra | Aug 2026 | ~1M† | ~$2.00 | ~$12 | Balanced everyday performance at a lower cost than Sol. Cached input ~$0.20 / 1M. |
| GPT-5.6 Luna | Aug 2026 | ~1M† | ~$0.20 | ~$1.20 | Fast and lightweight; optimized for high-volume, simpler tasks. Cached input ~$0.02 / 1M. |
| GPT-5.5 Instant | Jun 2026 | ~1M† | ~$5.00‡ | ~$30‡ | OpenAI's latest model; improvements to speed and instruction following. Cached input ~$0.50 / 1M. |
| GPT-5.4 Instant | Mar 2026 | ~1M† | ~$2.50 | ~$15 | Low-latency responses; large context window. Cached input ~$0.25 / 1M. |
| GPT-5.2 Instant | Oct 2025 | ~400K† | ~$1.75 | ~$14 | Low-latency variant of GPT-5.2 for faster responses. Cached input ~$0.175 / 1M. |
| GPT-5.2 Thinking | Oct 2025 | ~400K† | ~$1.75 | ~$14 | Strong reasoning model for coding and agentic tasks. Cached input ~$0.175 / 1M. |
| GPT-5 Mini | Aug 2025 | ~400K† | ~$0.25 | ~$2 | Cost-efficient; good for well-defined tasks at lower cost. Cached input ~$0.025 / 1M. |
| GPT-5 Nano | Aug 2025 | ~400K† | ~$0.05 | ~$0.40 | Ultra-lightweight; great for summarization and classification workloads. Cached input ~$0.005 / 1M. |
Google Model Options
| Model | Release Date | Context Window | Input $ / 1M | Output $ / 1M | Notes |
|---|---|---|---|---|---|
| Gemini 3.5 Flash (Low) | Recently added | ~1M† | ~$1.50‡ | ~$9.00‡ | Frontier-level intelligence at high speed and low cost; low thinking budget. Cached input ~$0.15 / 1M. |
| Gemini 3.5 Flash (High) | Recently added | ~1M† | ~$1.50‡ | ~$9.00‡ | Frontier-level intelligence at high speed and low cost; high thinking budget. Cached input ~$0.15 / 1M. |
| Gemini 3.1 Pro (Low) | Feb 2026 | ~1M† | ~$2.00 | ~$12.00 | Google's latest and most capable model; low thinking budget. Cached input ~$0.20 / 1M. |
| Gemini 3.1 Pro (High) | Feb 2026 | ~1M† | ~$2.00 | ~$12.00 | Higher reasoning effort variant of Gemini 3.1 Pro. Cached input ~$0.20 / 1M. |
| Gemini 3.1 Flash Lite (Low) | Feb 2026 | ~1M† | ~$0.25 | ~$1.50 | Google's lightest model for high-throughput, low-cost tasks; low thinking budget. Cached input ~$0.03 / 1M. |
| Gemini 3.1 Flash Lite (High) | Feb 2026 | ~1M† | ~$0.25 | ~$1.50 | Higher reasoning effort variant of Gemini 3.1 Flash Lite. Cached input ~$0.03 / 1M. |
| Gemini 3 Flash (Low) | Dec 2025 | ~1M† | ~$0.50 | ~$3.00 | Fast, efficient Google model; low thinking budget. Cached input ~$0.05 / 1M. |
| Gemini 3 Flash (High) | Dec 2025 | ~1M† | ~$0.50 | ~$3.00 | Higher reasoning effort Flash variant. Cached input ~$0.05 / 1M. |
Fireworks AI Models
| Model | Release Date | Context Window | Input $ / 1M | Output $ / 1M | Notes |
|---|---|---|---|---|---|
| Kimi 3 | Recently added | ~1M | ~$3.00 | ~$15.00 | Open-weight frontier model with a 1M-token context window and image input. Cached input ~$0.30 / 1M. |
| Kimi 3 (US) | Recently added | ~1M | ~$4.50 | ~$22.50 | Kimi 3 pinned to US-hosted inference with zero data retention; a 50% premium over Kimi 3. Price updated 2026-09-23. Cached input ~$0.45 / 1M. |
| GLM-5.3 | Sep 2026 | ~1M | ~$1.40 | ~$4.40 | Open-weight frontier model for general reasoning. Cached input ~$0.26 / 1M. |
| GLM-5.3 Flash | Sep 2026 | ~1M | ~$0.15 | ~$0.50 | Fast, low-cost multimodal model with image input. Cached input ~$0.03 / 1M. |
| MiniMax M3 | Jun 2026 | ~512K | ~$0.30 | ~$1.20 | Fast, cost-efficient multimodal model with image input. Cached input ~$0.06 / 1M. |
| DeepSeek V4.1 Flash | Recently added | ~1M | ~$0.30 | ~$1.20 | Fast, low-cost multimodal model with native image input. Price updated 2026-09-23. Cached input ~$0.006 / 1M. |
| Qwen3.8 Max | Recently added | ~262K | ~$2.00 | ~$6.00 | Frontier-scale model for complex reasoning and tool use. Cached input ~$0.25 / 1M. |
| Nemotron Lightning 3.5 | Sep 2026 | ~262K | ~$0.05 | ~$0.20 | Compact, low-cost open model for tool use. Cached input ~$0.01 / 1M. |
Footnotes:
- † Context window is provider-documented; it is not set in NinjaCat's model configuration. Values shown reflect the provider's published window and are pending confirmation.
- ‡ Customer rate card not yet published for this model; the value shown is carried over from the previous article and is pending confirmation before it should be relied upon.
Retired Fireworks Models
Some older Fireworks models have been retired (by Fireworks on 2026-09-25). Agents already using a retired model continue to work — requests are automatically routed to a current replacement model at request time, with no action required from you. Affected agents are billed at the replacement model's rate. Retired models are labeled (retired) in the model picker and can no longer be selected for new agents.
| Retired model | Now runs on |
|---|---|
| GLM-5.2 | GLM-5.3 |
| Kimi K2.7 Code | GLM-5.3 |
| DeepSeek V4 Pro | DeepSeek V4.1 Flash |
| DeepSeek V4 Flash | DeepSeek V4.1 Flash |
| Muse Glimmer 30B | Nemotron Lightning 3.5 |
Note: Qwen 3.7 Plus has been fully removed from the picker and is no longer available. Affected agents now use DeepSeek V4.1 Flash.
Note: In the Agent Builder, some models offer both a standard and a "Thinking" variant. The Thinking variant supports deeper reasoning but may come with higher cost and latency. For most agents, the standard variant is recommended unless your use case requires complex multi-step reasoning.
How to Choose the Right Model
AI models are continuously improving — what is "best" today may be surpassed in weeks or months. NinjaCat will continue evaluating and adding models that demonstrate better intelligence, efficiency, or performance.
General guidance:
- For most agents: Start with NinjaCat (US) — the default for new agents. NinjaCat selects and maintains a strong recommended model for you. You can switch at any time.
- For the most complex, high-effort, or long-horizon tasks: Claude Opus 5.5 or Claude Opus 5 — Anthropic's highest-capability models — or GPT-6 Astra (higher cost; reserve for high-value work).
- For complex reasoning or coding tasks: Claude Opus 4.7 or later, GPT-6.1 Sol, GPT-5.6 Sol, GLM-5.3, or Qwen3.8 Max.
- For balanced everyday performance: Claude Sonnet 5.5 or GPT-5.6 Terra are good middle-ground options.
- For speed or cost-sensitive tasks: GPT-6 Luna and Nemotron Lightning 3.5 (among the most cost-efficient options), Claude Haiku 4.5, GPT-5 Nano, Gemini 3.1 Flash Lite, or Gemini 3 Flash.
- For large context windows: OpenAI GPT-5.4 / GPT-5.6 series (~1M) or Google Gemini series (~1M). Note that the GPT-6 series uses a 272K standard context tier.
You can switch any agent's model at any time from the Agent Builder. If you switch, review your agent's instructions to confirm the prompt still performs as expected — some prompt adjustments may be needed when moving between models.
For the latest information from each provider, see their documentation:
Anthropic Claude Models OpenAI GPT Models Google Gemini
Note: When switching between models, prompt adjustments may be required to maintain optimal Agent performance. We will provide further guidance on prompt modifications as we continue testing and learning.
Updated 16 days ago
Did this page help you?