1M-Token Shared Context
Process large repositories, technical specifications, long document sets, and extended agent histories within a 1,048,576-token input-and-output window.
Kimi K3.1 Pro
Access Moonshot AI's Kimi K3.1 Pro through GPTProto for repository-scale coding, visual-to-code tasks, financial research, and long-running agents. Use one GPTProto key and balance to test it alongside 200+ text, image, video, and audio models.
API Usage Examples
curl --request POST "https://gptproto.com/v1/chat/completions" \
--header "Authorization: Bearer $GPTPROTO_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"model": "kimi-k3.1-pro",
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}'Kimi K3.1 Pro pricing
Chat, coding agents & document work. Priced per 1M tokens — input, cached input and output are billed separately.
| Scenario | MoonshotAI list | OpenRouter | GPTProto | You save / mo |
|---|---|---|---|---|
| Personal10M tokens / mo (4.8M cached) | $27.36 | $28.86 | $27.36 | −$0.00≈ $0.00 / yr |
| Team100M tokens / mo (48M cached) | $273.60 | $288.65 | $273.60 | −$0.00≈ $0.00 / yr |
| Business500M tokens / mo (240M cached) | $1368.00 | $1443.24 | $1368.00 | −$0.00≈ $0.00 / yr |
Save $0.00 / month
Annual savings ≈ $0.00 · pay as you go
Related Models
All Models
Kimi K3.1 ProCurrent
MoonshotAI · 1.05M context · $2/M input · $10/M output
Bytedance · 262K context · $0.0971/M input · $0.9714/M output
Bytedance · 262K context · $0.0971/M input · $0.9714/M output
Model
Price: GPTProto
vs Official
vs OpenRouter
Context
Stability
Kimi K3.1 ProCurrent
MoonshotAI
$2.00 / $10.00per 1M tokens
—
-5%
1.05M
Kimi K3.1 Pro API for Coding and Long-Horizon Agents
Kimi K3.1 Pro combines a 1,048,576-token context window with native visual understanding, always-on reasoning, tool calling, and unusually long text output. Moonshot reports a 65.7% score on Terminal-Bench 4.0 and a 61.2% score on its internal Kimi Code Bench 3.0 Long-horizon evaluation. GPTProto provides third-party API access through the same account used for other supported models.
Long Output for Multi-Step Work
The API defaults to 131,072 maximum output tokens and accepts a higher limit up to the shared 1M-token ceiling when the remaining context allows it.
Reported Coding Performance
Moonshot reports 65.7% on Terminal-Bench 4.0 and 61.2% on Kimi Code Bench 3.0 Long-horizon. Results depend on the harness, tools, and reasoning settings.
Reasoning and Agent Controls
Reasoning stays enabled with low, high, or max effort. The model also supports streaming, tool choice, structured output, and multi-step tool calls.
What Is the Kimi K3.1 Pro API?
Kimi K3.1 Pro is the first model in Moonshot AI's K3.1 series. It is a post-trained upgrade to Kimi K3, using the same 3-trillion-class foundation with more reinforcement-learning compute across coding, agentic work, finance, and 3D creation. The API accepts text and image input and returns text. Text-to-text applications can send chat messages, documents, code, and tool results without using the vision path.
The model uses Preserved Thinking: reasoning is always on, and multi-turn applications should return the complete assistant message, including reasoning_content, on the next request. Developers can select low, high, or max reasoning effort. These constraints matter when migrating an existing Kimi integration.
| Specification | Kimi K3.1 Pro |
|---|---|
| Developer | Moonshot AI |
| Model class | 3T-class post-trained reasoning model |
| Official model ID | kimi-k3.1-pro |
| Input to output | Text and images to text |
| Shared context window | 1,048,576 tokens |
| Default maximum output | 131,072 tokens |
| Configurable output ceiling | Up to 1,048,576 tokens, subject to the shared context limit |
| Reasoning | Always on; low, high, or max (default) |
| Tool choice | auto, none, or required |
| Streaming and structured output | Supported |
| Open-weight status | API live; full weights scheduled for release by November 4, 2026 |
Kimi K3.1 Pro API Applications
Repository-scale coding. Supply source files, interfaces, tests, and configuration. Ask the model to trace dependencies before changing code, then require a patch and test plan. The large context window reduces the need to split connected modules into unrelated prompts.
Long-horizon agentic work. Use function calling and streamed tool arguments for agents that inspect files, search documentation, run commands, and revise a plan. Keep permissions, timeouts, and acceptance tests in the surrounding agent.
Financial research. Supply filings, spreadsheet extracts, calculation rules, and source notes, then request a traceable analysis with evidence and explicit assumptions. Validate every material figure before use.
Visual-to-code and 3D workflows. Use screenshots, sketches, or reference images for frontend implementation, Three.js scenes, Blender scripts, and CAD tasks. Run and inspect generated code and geometry.
Long technical output. Generate migration plans, large patches, or multi-file drafts. Set an output budget because input and generated output share the same 1M-token window.
Kimi K3.1 Pro vs Kimi K3
K3.1 Pro is not a larger-context successor to K3; both provide the same 1M-token window and native visual understanding. The upgrade is mainly in post-training, measured task performance, and launch pricing.
| Decision factor | Kimi K3.1 Pro | Kimi K3 |
|---|---|---|
| Position | Current K3.1 flagship | Previous-generation K3 flagship |
| Base and training | K3-based model with substantially more post-training compute | Original 2.8T K3 base model |
| Context | 1,048,576 tokens | 1,048,576 tokens |
| Vision input | Supported | Supported |
| Kimi Code Bench 3.0 Long-horizon | 61.2% | 55.95% |
| Moonshot list price at launch | $2 input / $10 output per 1M tokens | $3 input / $15 output per 1M tokens |
| Open weights | Scheduled by November 4, 2026 | Released |
| Best fit | New coding, research, and long-running agent deployments | Existing K3 deployments that do not yet need the upgrade |
Benchmark values above are reported by Moonshot AI and were not independently reproduced by GPTProto. Run both models with the same prompts, tools, token limits, and acceptance tests before changing production traffic.
Kimi K3.1 Pro vs Other Coding and Agent Models
Compare API behavior and workload fit rather than relying on one leaderboard rank.
| Model | What distinguishes it | When to choose it |
|---|---|---|
| Kimi K3.1 Pro | 1M shared context, long decode, native vision, always-on reasoning, and an open-weight path | Coding agents, large-context research, visual-to-code, or future self-hosting |
| DeepSeek V4.1 Flash | Lower token rates and a Flash deployment profile | High-volume work where cost and throughput come first |
| Gemini 4 Argon | Text, image, audio, video, and PDF input in Google's ecosystem | Broad multimodal applications or Gemini-centered teams |
| GPT-6 Astra | Computer use and OpenAI's Codex and Responses ecosystem | Screen-based agents and computer-use workflows |
| Claude Opus 5.5 | Anthropic reports 66.4% on Terminal-Bench 4.0 | Agentic coding when project evals favor Anthropic's tool behavior |
Provider benchmarks use different harnesses and settings. Compare candidates in the GPTProto playground, then grade task completion, invalid tool calls, latency, output use, and total cost on the same workload.
Migration Checks Before You Upgrade from Kimi K3
Kimi K3 and K3.1 Pro share request-parameter constraints, so most K3 integrations only need the model selection changed to the identifier shown in GPTProto's live API Usage tab.
Do not send the K2-style
thinkingparameter. K3-series reasoning is always enabled.Set top-level
reasoning_efforttolow,high, ormax;maxis the default.Return the full assistant message, including
reasoning_content, during multi-turn chats and tool loops.Keep supported sampling defaults and recheck output budgeting inside the shared context window.
Run a canary on real repositories and tool schemas before routing all traffic to the new model.
When Should You Choose Kimi K3.1 Pro?
Choose Kimi K3.1 Pro for codebase migrations, research agents, financial analysis, CAD scripting, or other work that needs large context, visual references, and repeated tool use.
Choose another model when the priority is computer control, audio or video input, the lowest token rate, or a provider-specific tool ecosystem. GPTProto lets the application test supported alternatives with one key and balance.
Kimi K3.1 Pro API Frequently Asked Questions
How much does the Kimi K3.1 Pro API cost on GPTProto?
At page launch, GPTProto uses Kimi's list rate: $2 per 1M input tokens, $10 per 1M output tokens, $2 per 1M cache-write tokens, and $0.20 per 1M cache-hit tokens. The live panel remains the billing source of truth. See GPTProto pricing.
How do I get a Kimi K3.1 Pro API key and access the playground?
Create an account and GPTProto API key, then use Playground for testing. Copy the current endpoint and model identifier from API Usage. The same key can access other supported models.
What are the Kimi K3.1 Pro context window and output limit?
The API has a shared 1,048,576-token context window. max_tokens defaults to 131,072 and can be configured up to 1,048,576, but the combined input and generated output must remain within the shared window.
Is Kimi K3.1 Pro a text-to-text API?
Yes. It accepts text prompts and returns text for chat, coding, extraction, research, and agent workflows. The underlying model also accepts image input, so calling it text-only would understate its capabilities.
Can reasoning be disabled in the Kimi K3.1 Pro API?
No. Preserved Thinking is always enabled. Use reasoning_effort: "low", "high", or "max" to control the reasoning budget instead of sending a disabled thinking mode.
Does Kimi K3.1 Pro support tool calling and structured output?
Yes. It supports tool_choice values of auto, none, and required, streamed tool arguments, and JSON-schema structured output. Keep execution and validation in the application layer.
Is Kimi K3.1 Pro better than Kimi K3 for coding?
Moonshot reports 61.2% for K3.1 Pro versus 55.95% for K3 on Kimi Code Bench 3.0 Long-horizon. Confirm the improvement on your own repositories and agent harness.
Does GPTProto provide private Kimi K3.1 Pro API access?
GPTProto uses account-scoped API keys rather than a public chat. For Zero Data Retention, regional processing, or contractual privacy terms, confirm GPTProto's applicable policy with support before sending regulated data.
Is Kimi K3.1 Pro open weight?
Moonshot launched the API first and scheduled full model weights for release by November 4, 2026. Treat self-hosting requirements as pending until publication.
Related Articles
Guides, comparisons, and updates related to this model.
All Articles
How to Get Kimi K3 API Access: Setup, Pricing, and Real Use Cases
Get Kimi K3 API access on GPTProto. Copy the setup guide to your AI coding assistant, explore token pricing, and build agents or SaaS features.
Kimi K3 vs Claude Opus 5.5: Which Is Better for Coding and AI Agents?
Compare Kimi K3 and Claude Opus 5.5 on coding, frontend work, agent benchmarks and API cost. See which model is better for your workload.
Best AI for Roleplay in 2026: 6 Models Compared for Memory, Stories, and RPGs
Compare the best AI models for roleplay in 2026 for memory, story writing, RPGs and long chats, including Kimi K3, GLM, Claude, DeepSeek and Gemini.
Best Uncensored LLM APIs in 2026: Which Should You Choose for Roleplay?
Compare Grok, GLM, DeepSeek, Hermes and Dolphin for roleplay apps. See API access, refusal limits and long-chat costs before choosing your model.