Kimi K2.5
Kimi K2.5 is worth evaluating for coding, rag, and agents when its provider route and context window match the workload.
Use it for
- Teams evaluating coding, rag, and agents
- Workloads that can use a 256k context window
- Buyers comparing 4 tracked provider routes
Do not use it for
- Workloads where another current model has stronger sourced task evidence
- Family
- Kimi
- Released
- 2026-03-15
- Context
- 256k
- Parameters
- 1T (MoE, 384 experts)
- Architecture
- Mixture of Experts
- Specialization
- code
- Openness
- Proprietary
- License
- ProprietaryCommercial use: conditional
- Weights
- Not released
- Code
- Unknown
- Training
- Fine-tuned
Cheapest of 10 routes · OpenRouter
About
Kimi K2.5 is Moonshot AI's Kimi model focused on code generation and software engineering. It offers a 256K-token context window and scores 87.9 on GPQA.
Kimi K2.5 is a proprietary model in the Kimi family. The structured metadata tracks a 256k-token context window, multimodal input, function calling, and structured outputs. This page tracks provider routes through Cloudflare Workers AI, Fireworks AI, OpenRouter, and 7 more, with the cheapest tracked route listed at $0.44 input and $2 output per 1M tokens. Headline tracked benchmarks include Google-Proof Q&A 87.9, MMLU PRO 87.1, and BFCL 47.1.
Top use-case fit: coding, agents, and build tasks
Coding
Q/$ C2 relevant benchmarks in the decision map.
RAG
Included by capability and metadata signals in the decision map.
Agents
Q/$ C4 relevant benchmarks in the decision map.
Provider price ladder
Compare all 10Compare API pricing across 4 providers for input and output tokens, batch, and cached reads when available.
| Provider | Input / 1M | Output / 1M | Route |
|---|---|---|---|
| OpenRouter | $0.440 | $2.00 | Serverless |
| Together AI | $0.500 | $2.80 | Serverless |
| AWS Bedrock | $0.600 | $3.00 | Serverless |
| Fireworks AI | $0.600 | $3.00 | Serverless |
Available via routers & gateways(8)
LiteLLM
GatewayOpen-source Python SDK and proxy server that unifies 100+ LLM APIs behind a single OpenAI-compatible interface, with load balancing, cost tracking, and configurable failover.
OpenRouter
HybridUnified hybrid gateway to 400+ models from 60+ providers via a single OpenAI-compatible API, with optional auto-routing that selects the best model per prompt.
Portkey
GatewayProduction AI gateway routing to 1,600+ LLMs with failover, load balancing, semantic caching, and guardrails; Apache 2.0 core is fully self-hostable with the complete feature set.
Amazon Bedrock Intelligent Prompt Routing
RouterAWS Bedrock's native intelligent prompt router that routes prompts between Anthropic Claude model tiers (Haiku/Sonnet) based on predicted task complexity, with no extra per-routing charge.
Azure AI Foundry Model Router
RouterMicrosoft Azure AI Foundry's native model router that uses a trained ML model to route each prompt in real time to the optimal Azure-hosted model, with Balanced/Cost/Quality mode selection and automatic failover.
Helicone
GatewayObservability-first AI gateway with routing, caching, rate limiting, and request tracing; Apache 2.0 open-source core with a managed hosted tier for logging and analytics.
Capabilities
Benchmark peer barsfor Coding
Benchmark scores(17)
| Benchmark | Score | Version | Evaluation | Source |
|---|---|---|---|---|
| Google-Proof Q&A | 87.9 | diamondObserved 2026-04-18 | — | Source |
| MMLU PRO | 87.1 | Thinking mode (accuracy)Observed 2026-06-07 | — | Source |
| BFCL | 47.1 | v4Observed 2026-04-19 | — | Source |
| τ-bench | 74.2 | τ-benchObserved 2026-04-24 | — | Source |
| MultiChallenge | 61.4 | MultiChallengeObserved 2026-04-26 | — | Source |
| MMMU Pro | 78.5 | LLM-Stats aggregatorObserved 2026-06-07 | — | Source |
| SWE-rebench | 58.5 | pass@1 (best of 5 runs)Observed 2026-05-28 | — | Source |
| AIME 2025 | 96.1 | Thinking mode (accuracy)Observed 2026-06-07 | — | Source |
| Berkeley Function Calling Leaderboard v3 | 64.5 | BFCL v3 (accuracy%)Observed 2026-06-07 | — | Source |
| BrowseComp | 60.6 | BrowseComp (accuracy%)Observed 2026-06-07 | — | Source |
| Humanity's Last Exam | 50.2 | HLE-Full with tools (agentic) (accuracy)Observed 2026-06-07 | — | Source |
| LiveCodeBench | 85.0 | LiveCodeBench v6 (pass@1)Observed 2026-06-07 | — | Source |
| MATH-500 | 98.0 | Thinking mode (accuracy)Observed 2026-06-07 | — | Source |
| MCP-Atlas | 29.5 | MCP-Atlas (accuracy%)Observed 2026-06-07 | — | Source |
| SWE-bench Verified | 76.8 | From official GitHub model card (resolved)Observed 2026-06-07 | — | Source |
| Terminal-Bench 2.0 | 50.8 | Terminal-Bench 2.0 (accuracy%)Observed 2026-06-07 | — | Source |
| CursorBench | 31.9 | CursorBench 3.1Observed 2026-06-30 | Configuration: Kimi 2.5 (single reported configuration) Harness: CursorBench 3.1 Evaluator: Cursor Confidence: confirmed Notes: Cursor published one CursorBench 3.1 configuration for this model; no cross-effort selection was needed. | Source |
Migration checks
No linked migration route is available for this model yet.
Rankings & picks(2)
Compare Kimi K2.5 with other models
Comparison and alternatives
Browse all comparisons →Show all 65 popular comparisonssorted by 7-day search impressions
Frequently asked questions
What is the context window of Kimi K2.5?
Kimi K2.5 has a context window of 256k tokens.
How much does Kimi K2.5 cost?
Kimi K2.5 pricing ranges from $0.44/1M to $0.6/1M input tokens depending on the provider.
When was Kimi K2.5 released?
Kimi K2.5 was released on 2026-03-15.
Which providers offer Kimi K2.5?
Kimi K2.5 is available from 10 providers: Cloudflare Workers AI, Fireworks AI, OpenRouter, Together AI, NVIDIA NIM, AWS Bedrock, Replicate API, Microsoft Foundry, Vercel AI Gateway, Novita AI.
What benchmarks has Kimi K2.5 been tested on?
Kimi K2.5 has been evaluated on 17 benchmarks, including Google-Proof Q&A, MMLU PRO, BFCL, τ-bench, MultiChallenge.
Cheapest of 10 routes · OpenRouter