Models & Labs
Kimi K3 Challenges Claude Fable 5 on DeepSWE
Together AI Bloghigh confidence
Why it matters
- →Kimi K3 offers a cost-effective alternative to Claude Fable 5, making it accessible for more teams.
- →Its open-weight nature allows for customizable deployment, enhancing flexibility.
- →The model's strong performance in multi-attempt scenarios makes it suitable for retry-tolerant tasks.
Kimi K3, an open-weight model, is challenging Claude Fable 5 on the DeepSWE benchmark, a test of software engineering capabilities. While Claude Fable 5 leads slightly in single-attempt success rates, Kimi K3 excels in multi-attempt scenarios, offering better pass@2 and pass@4 results. Kimi K3 is also significantly cheaper, costing about a third of Claude Fable 5's price per rollout. This cost-effectiveness, combined with its open-weight nature, makes Kimi K3 an attractive option for teams seeking flexibility and efficiency in AI deployment.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
More in Models & Labs
Models & Labsother
llama.cpp b12340 adds CUDA 13 and ROCm 10.0
This release quietly closes the hardware gap for local inference by adding default support for CUDA 13 and ROCm 10.0 across Linux and Windows. Users with newer NVIDIA GPUs or AMD accelerators no longer need to compile from source to get performance, while Apple Silicon builds now ship KleidiAI disabled by default, shifting the focus back to standard ARM optimizations. The inclusion of Snapdragon support for Linux and Android marks a significant step toward on-device AI beyond the x86/ARM desktop ecosystem. It is less about new features and more about ensuring llama.cpp remains the universal runtime for every available accelerator.
llama.cpp Releases
Models & Labscoding
llama.cpp b12342 adds few-row MMQ for speculative decoding
Speculative decoding just got a significant speed boost on NVIDIA GPUs. This release introduces a specialized kernel for 'few-row' matrix multiplications, targeting the small batch sizes typical of verification passes in speculative decoding workflows. By optimizing these specific operations for quantized formats like Q8_0 and IQ4 variants, llama.cpp reduces latency where it matters most during token generation. The change is CUDA-only for now but leverages hardware capabilities from Ampere onwards, making local inference faster without requiring new model architectures or complex configuration changes.
