# Build powerful agents on OSS models with blazing fast inference on Fireworks

> See Artificial Analysis speed results for Kimi K2.5, DeepSeek V3.2, GLM 4.7, and GPT-OSS 120B on Fireworks, and how speculative decoding cuts latency.

Announcement

JAN 27, 2026

[Kimi K2.5](https://fireworks.ai/blog/kimi-k2p5) just dropped yesterday and is available Day 0 on Fireworks! As open models get more powerful and agentic, low latency enables complex, multi-step AI agents to be usable in real-time. **Fireworks is fastest across the top open models among GPU-based providers, as benchmarked by Artificial Analysis.** Get the top speed you need, tailored for your use case, only with Fireworks.

[**Kimi K2.5**](https://fireworks.ai/blog/kimi-k2p5)

![Kimi K2.5 Benchmarks](https://cdn.sanity.io/images/pv37i0yn/horizon/7783ef68df1d57611d1e4207ad57a86623cb8f9d-4984x1612.png)

Artificial Analysis Kimi K2.5 Chart

Stay tuned as our engineering team continues to optimize the performance! 

[**Kimi K2 Thinking**](https://fireworks.ai/models/fireworks/kimi-k2-thinking)

![Kimi K2 Thinking](https://cdn.sanity.io/images/pv37i0yn/horizon/e0d216ba5b880b954348d57bc5a08b8260a48722-3960x1644.png)

Artificial Analysis Kimi K2 Thinking Chart

[**Deepseek V3.2 Reasoning**](https://fireworks.ai/models/fireworks/deepseek-v3p2)

![Deepseek Benchmark Chart](https://cdn.sanity.io/images/pv37i0yn/horizon/ab7982dc4d3989d29de8ecd19b8fa76b36fb5c04-3960x1644.png)

Artificial Analysis Deepseek V3.2 Speed Chart

[**Deepseek V3.2 Non-Reasoning**](https://fireworks.ai/models/fireworks/deepseek-v3p2)

![Deepseek V3.2 Non-Reasoning](https://cdn.sanity.io/images/pv37i0yn/horizon/60b7b6e7ee1f11710a803b7a77f8769855a638d3-3960x1644.png)

Artificial Analysis Deepseek V3.2 Speed Chart

[**GLM 4.7 Reasoning**](https://fireworks.ai/models/fireworks/glm-4p7)

![GLM 4.7 Benchmark](https://cdn.sanity.io/images/pv37i0yn/horizon/5f58c772d4be0ddbc729d340271e411e51c28789-3960x1644.png)

Artificial Analysis GLM 4.7 Reasoning Chart

[**GPT-OSS 120B**](https://fireworks.ai/models/fireworks/gpt-oss-120b)

![GPT-OSS Benchmark Chart](https://cdn.sanity.io/images/pv37i0yn/horizon/4e667c4950ec9545aaace2031953180f8218d1a1-3960x1644.png)

Artificial Analysis GPT-OSS 120B Chart

## Unlocking Peak Performance: Fireworks Customization Engine and Infrastructure

Fireworks' customization engine and virtual cloud infrastructure are engineered to deliver best-in-class performance for developers. We've built the following advanced capabilities to enhance speed as seen in the benchmarks, and across multiple different usecases:

   - [**FireOptimizer: Intelligent Resource Management**](https://fireworks.ai/blog/3d-fireoptimizer)

FireOptimizer maximizes performance by optimizing three core dimensions:

   - **Deployment Shape:** Enables strategic GPU selection
   - **Sharding Strategy:** provides advanced parallelism techniques.
   - **Scheduling:** Expertly balancing Time-to-First-Token (TTFT) with overall throughput.

This ensures your hardware is precisely right-sized to meet the specific Service Level Agreements (SLAs) of your application.

   - **Speculative Decoding: Accelerate Latency-Sensitive Applications**

For applications where latency is critical, such as code assistance and agentic reasoning, Fireworks leverages speculative decoding. This process employs a smaller "draft" model to anticipate token sequences, which the larger "target" model then validates in parallel. This methodology achieves a significant 2x to 3x speedup in generation without sacrificing the superior output quality of the main model.

   - **Custom Kernels for Optimized Virtual Cloud Infrastructure**

To extract the maximum possible speed and efficiency from NVIDIA Blackwell GPU hardware, our engineering team developed customized kernels, ensuring optimal performance and fully leveraging the power of the underlying GPU architecture.

Together, these proprietary advancements ensure Fireworks delivers the unrivaled speed confirmed by independent Artificial Analysis benchmarks across the industry's best open-source models. To further improve performance, you can leverage our Fine Tuning platform, to bring your own data to tailor your workload to your usecase. Get started [today](https://app.fireworks.ai/signup) or [contact us](https://fireworks.ai/contact) for help with your application.

[Announcement](https://fireworks.ai/blog?blogTypes=Announcement)
