Models & Labs
DeepSeek V4 Pro 0813 vs GPT-5.6 Sol: Cost Efficiency
Together AI Bloghigh confidence
Why it matters
- →Demonstrates cost-effective AI model deployment strategies.
- →Highlights trade-offs between accuracy, speed, and cost in AI models.
- →Offers insights into optimizing AI model usage for software engineering tasks.
DeepSeek V4 Pro 0813 and GPT-5.6 Sol were compared on the DeepSWE benchmark, which evaluates software engineering capabilities. GPT-5.6 Sol demonstrated superior single-attempt accuracy and speed, solving 72.7% of tasks on the first try. However, DeepSeek V4 Pro 0813 proved to be 35 times cheaper, offering broader coverage over multiple attempts. A combined approach, using Pro first and escalating to Sol when necessary, achieved an 83% task completion rate at a lower cost than using Sol alone. This strategy highlights the cost-effectiveness of Pro while leveraging Sol's precision when needed.
Read originalThe story around this
TopicDeepSeek V4.1 And GLM 5.3
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
More from Together AI Blog
More in Models & Labs
Models & Labscoding
vLLM v0.31.0: SM100 defaults and fast restarts
vLLM is quietly becoming the definitive runtime for NVIDIA's latest hardware, making NVFP4 compressed KV caches the default for DeepSeek-V4.1-Flash on SM100 GPUs. This isn't just a performance tweak; it fundamentally changes how enterprise inference scales by keeping post-quantized weights resident in GPU memory across engine restarts via the new preload daemon. The release also hardens speculative decoding with Model Runner V2, fixing OOMs that previously plagued wide expert deployments. For builders, this means lower latency and higher throughput on next-gen hardware without manual configuration overhead.
vLLM Releases
Models & Labsmodels
llama.cpp v0.6.0 adds decision model API and Apple Silicon speedups
This release shifts llama.cpp from a pure text engine to a multimodal inference runtime capable of handling 'decision models' like Clef and GLM-5.3-Flash. The new /v1/systemone server endpoint standardizes how these non-autoregressive models are queried, while the extended batch API allows mixed token and embedding inputs for complex architectures. Apple Silicon users get a tangible performance boost with new Metal MMA kernels that accelerate speculative decoding by up to 3x. It’s a significant step toward supporting the next generation of hybrid reasoning models locally.
