PERFORMANCELlama 3.3 70B → Scaleway: throughput ↑ 53%6hPERFORMANCEGLM-5.2 → Scaleway: TTFT ↓ 30%6hPERFORMANCEMistral Medium 3.5 → Scaleway: throughput ↑ 22%6hPERFORMANCEQwen3-235B → Scaleway: throughput ↑ 67%6hPERFORMANCEGPT-5.6 Sol → OpenAI: TTFT ↓ 22%6hPERFORMANCEMistral Large 3 → Mistral AI: TTFT ↓ 21%6hPERFORMANCEGPT-5.6 Luna → OpenAI: throughput ↑ 21%6hPERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↑ 147%6hPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↓ 71%6hPERFORMANCEGPT-OSS 120B → Scaleway: TTFT ↓ 46%6hPERFORMANCEGPT-OSS 120B → Together AI: TTFT ↓ 40%6hPERFORMANCEKimi K3 → Together AI: TTFT ↓ 20%6hPERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↑ 36%30hPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↓ 19%30hPERFORMANCEQwen3-235B → Scaleway: throughput ↑ 19%2dPERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↑ 58%2dPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↑ 21%2dPERFORMANCEGPT-5.6 Sol → OpenAI: throughput ↑ 16%2dPERFORMANCEGPT-OSS 120B → Together AI: TTFT ↑ 22%2dPERFORMANCEKimi K3 → Together AI: throughput ↑ 62%2dPERFORMANCEKimi K3 → Together AI: TTFT ↓ 42%2dPERFORMANCEQwen3-235B → Scaleway: throughput ↑ 30%3dPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↓ 56%3dPERFORMANCELlama 3.3 70B → Scaleway: throughput ↑ 34%3dPERFORMANCELlama 3.3 70B → Scaleway: TTFT ↓ 26%3dPERFORMANCEKimi K3 → Together AI: throughput ↑ 128%3dPERFORMANCEGPT-OSS 120B → Together AI: throughput ↑ 29%3dPERFORMANCEGPT-OSS 120B → Together AI: TTFT ↓ 21%3dPERFORMANCEGPT-OSS 20B → Groq: TTFT ↑ 66%3dPERFORMANCEGLM-5.2 → Together AI: throughput ↑ 47%3dPERFORMANCEMistral Medium 3.5 → Scaleway: TTFT ↑ 18%4dPERFORMANCEGLM-5.2 → Scaleway: throughput ↓ 27%4dPERFORMANCEGLM-5.2 → Scaleway: TTFT ↑ 49%4dPERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↓ 71%4dPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↑ 190%4dPERFORMANCEGPT-5.6 Luna → OpenAI: throughput ↓ 23%4dPERFORMANCELlama 3.3 70B → Scaleway: TTFT ↑ 60%4dPERFORMANCEGLM-5.2 → Together AI: TTFT ↓ 25%4dPERFORMANCEGPT-OSS 120B → Together AI: throughput ↓ 16%4dPERFORMANCEGPT-OSS 120B → Together AI: TTFT ↓ 17%4d
ENGINEERING STUDY

Tool Calling Is an Infrastructure Property

The same model's function-calling reliability varies wildly with the provider serving it.

Key findings
  • On DeepSeek V4 Flash, measured tool-call reliability spans 0.0%–100.0% across 3 providers serving identical weights.
  • Scaleway leads the measured set on this model; the weakest path fails roughly 100 of every 100 tool calls.
  • Identical weights, different serving stacks, different tool-call outcomes: the split is infrastructure-side.

Where tool calls break

A failed tool call is rarely the model refusing — it is malformed JSON, truncated arguments, or a serving stack that mangles the function-calling protocol under load. That is infrastructure behaviour, and it only becomes visible when the same model is measured across providers under an identical harness.

ProviderTool reliability
Scaleway100.0%
Fireworks0.0%
Together0.0%
Measured tool-call reliability on DeepSeek V4 Flash (Agents profile) · ● MEASURED
What this means for engineers
  • For agent stacks, provider selection is a first-order reliability decision — test the exact path, not the model in the abstract.
Dataset

Figures recompute from the live rolling window on every visit; the snapshot and versions above are the provenance of the regime that produced them. View the current benchmark →

Cite this researchInferenceBench (2026). Tool Calling Is an Infrastructure Property. InferenceBench Research.
← All research
Related research
Tool Calling Is an Infrastructure Property · InferenceBench