OVHcloud AI Endpoints - Model Selection Guide¶

Last Updated: 2026-06-03 Context: OpenCode with OVHcloud AI Endpoints integration

Getting Started with OpenCode¶

Connect OpenCode to OVH Run the connect command inside OpenCode:

/connect

Search for “OVHcloud AI Endpoints” and enter your API key when prompted.

Select a model List available models with /models and choose the gpt-oss-120b model (ovhcloud/gpt-oss-120b). This model offers strong reasoning capabilities at a low cost.
Test the setup Try a simple query, e.g., “What is 2+2?” or ask OpenCode to generate a small code snippet.

Quick Recommendations¶

Best value: gpt-oss-120b – 117B parameters, fp4 quantization, excellent reasoning and responsiveness.
Fast & cheap alternative: gpt-oss-20b – 21B parameters, very low latency.
High‑performance option: qwen3.6-27b – 35B parameters, multimodal support.

Models Comparable to Claude Sonnet for Interactive Chat¶

Based on the OVHcloud catalog, here are the models suitable for interactive chat that are comparable to Claude Sonnet.

Top Recommendations (Most Comparable to Claude Sonnet)¶

1. Qwen3.6-27B (Visual LLM) ⭐ NEW¶

Parameters: 35B (fp8 quantization)
Context: 262K tokens
Capabilities: Function calling, Multimodal, Reasoning
Pricing: €0.4/Mtoken input, €2.7/Mtoken output
Model ID: ovhcloud/qwen3.6-27b
Responsiveness: ⚡⚡⚡⚡ Excellent (27B model with fp8 quantization provides fast inference)
Best For: General interactive chat, coding assistance, multimodal tasks with good speed/quality balance

2. Qwen3.5-397B-A17B (Visual LLM) ⭐ NEW - MOST POWERFUL¶

Parameters: 397B (fp8 quantization)
Context: 262K tokens
Capabilities: Function calling, Multimodal, Reasoning
Pricing: €0.6/Mtoken input, €3.6/Mtoken output
Model ID: ovhcloud/qwen3.5-397b-a17b
Responsiveness: ⚡⚡⚡ Good (larger model, slightly slower but still production-ready)
Best For: Complex reasoning tasks, multi-file refactoring, architectural decisions

3. gpt-oss-120b (Reasoning LLM) 💰 BEST VALUE¶

Parameters: 117B (fp4 quantization)
Context: 131K tokens
Capabilities: Function calling, Reasoning
Pricing: €0.08/Mtoken input, €0.4/Mtoken output
Model ID: ovhcloud/gpt-oss-120b
Responsiveness: ⚡⚡⚡⚡ Excellent (fp4 quantization speeds up inference significantly)
Best For: Cost-effective coding assistance with strong reasoning capabilities

Good Alternatives¶

4. Qwen3-32B (Reasoning LLM)¶

Parameters: 32.8B (fp8 quantization)
Context: 32K tokens
Capabilities: Function calling, Reasoning
Pricing: €0.08/Mtoken input, €0.23/Mtoken output
Model ID: ovhcloud/qwen3-32b
Responsiveness: ⚡⚡⚡⚡⚡ Very Fast (smaller model, fp8 quantization)
Best For: Quick iterations, fast responses for simple to medium complexity tasks

5. Meta-Llama-3.3-70B-Instruct (LLM)¶

Parameters: 70B (fp8 quantization)
Context: 131K tokens
Capabilities: Function calling
Pricing: €0.67/Mtoken input, €0.67/Mtoken output
Model ID: ovhcloud/meta-llama-3_3-70b-instruct
Responsiveness: ⚡⚡⚡⚡ Excellent (well-optimized, balanced performance)
Best For: General-purpose coding, documentation, established Llama ecosystem familiarity

6. Mistral-Small-3.2-24B-Instruct-2506 (Visual LLM)¶

Parameters: 24B (bf16 quantization)
Context: 128K tokens
Capabilities: Function calling, Multimodal
Pricing: €0.09/Mtoken input, €0.28/Mtoken output
Model ID: ovhcloud/mistral-small-3.2-24b-instruct-2506
Responsiveness: ⚡⚡⚡⚡ Excellent (smaller size, good optimization)
Best For: European data sovereignty preference, Mistral ecosystem users

Budget-Friendly Options¶

7. Qwen3.5-9B ⭐ NEW - FASTEST¶

Parameters: 9.7B (bf16 quantization)
Context: 262K tokens
Capabilities: Function calling, Multimodal, Reasoning
Pricing: €0.1/Mtoken input, €0.15/Mtoken output
Model ID: ovhcloud/qwen3.5-9b
Responsiveness: ⚡⚡⚡⚡⚡ Very Fast (small model size = near-instant responses)
Best For: Rapid prototyping, simple edits, learning, budget-conscious development

8. gpt-oss-20b (Reasoning LLM) 💰 CHEAPEST¶

Parameters: 21B (fp4 quantization)
Context: 131K tokens
Capabilities: Function calling, Reasoning
Pricing: €0.04/Mtoken input, €0.15/Mtoken output
Model ID: ovhcloud/gpt-oss-20b
Responsiveness: ⚡⚡⚡⚡⚡ Very Fast (fp4 quantization, small model)
Best For: High-volume tasks, experimentation, CI/CD integration

9. Qwen3-Coder-30B-A3B-Instruct (Code LLM) 🔧 CODE SPECIALIST¶

Parameters: 30B (fp8 quantization)
Context: 256K tokens
Capabilities: Function calling, Code Assistant
Pricing: €0.06/Mtoken input, €0.22/Mtoken output
Model ID: ovhcloud/qwen3-coder-30b-a3b-instruct
Responsiveness: ⚡⚡⚡⚡ Excellent (optimized specifically for code generation)
Best For: Pure coding tasks, code completion, refactoring, bug fixes

Responsiveness Comparison Table¶

Model	Size	Quantization	Speed Rating	Latency (Est.)*	Use Case
Qwen3.5-9B	9.7B	bf16	⚡⚡⚡⚡⚡	~100-200ms	Fastest responses
gpt-oss-20b	21B	fp4	⚡⚡⚡⚡⚡	~150-250ms	Fast & cheap
Qwen3.6-27B	35B	fp8	⚡⚡⚡⚡	~200-400ms	Balanced choice
Qwen3-Coder-30B	30B	fp8	⚡⚡⚡⚡	~200-400ms	Code-focused
Qwen3-32B	32.8B	fp8	⚡⚡⚡⚡	~250-450ms	Quick reasoning
Meta-Llama-3.3-70B	70B	fp8	⚡⚡⚡⚡	~300-600ms	Reliable workhorse
gpt-oss-120b	117B	fp4	⚡⚡⚡⚡	~400-700ms	Strong reasoning
Qwen3.5-397B	397B	fp8	⚡⚡⚡	~800-1500ms	Most powerful

*Estimated latency for typical prompt (500 input tokens, 100 output tokens) on OVH EU infrastructure. Actual latency depends on server load, network conditions, and prompt complexity.

Key Capabilities Comparison¶

Feature	Claude Sonnet 4.5	OVH Top Models	Notes
Function Calling	✅ Yes	✅ Yes (all recommended)	Essential for OpenCode tool use
Extended Thinking	✅ Yes	✅ Yes (Qwen/gpt-oss series)	Advanced reasoning capabilities
Multimodal (Vision)	✅ Yes	✅ Yes (Qwen 3.5/3.6, Mistral-Small)	Image understanding
Context Length	200K	🏆 Up to 262K (Qwen3.5/3.6)	OVH wins on context
Parameters	~200B+	🏆 Up to 397B (Qwen3.5-397B)	Comparable scale
Responsiveness	⚡⚡⚡⚡⚡	⚡⚡⚡⚡ to ⚡⚡⚡⚡⚡	Varies by model size
Data Sovereignty	❌ US-based	✅ European (OVH)	GDPR compliance
Pricing	$$$ Premium	$ to $$ Competitive	Significant cost savings

Quantization Impact on Responsiveness¶

Understanding quantization helps predict model responsiveness:

fp4 (4-bit): 🚀 Fastest inference, slight quality trade-off
Examples: gpt-oss-120b, gpt-oss-20b
~4x faster than fp16, ~2x faster than fp8
fp8 (8-bit): ⚡ Fast inference, minimal quality loss
Examples: Qwen3.6-27B, Qwen3.5-397B, Meta-Llama-3.3-70B
~2x faster than fp16, excellent quality/speed balance
bf16 (16-bit): 💎 Full quality, slower inference
Examples: Qwen3.5-9B, Mistral-Small-3.2-24B
Best quality but needs smaller models for good responsiveness

Recommendations by Use Case¶

Best Overall for Interactive Chat 🏆¶

Qwen3.6-27B - Perfect balance of speed, quality, and features

Fast enough for interactive use (⚡⚡⚡⚡)
Strong reasoning and multimodal capabilities
Large context window (262K)
Reasonable pricing

Best for Speed & Cost 💰¶

gpt-oss-120b - Excellent value proposition

Very responsive (fp4 quantization)
Strong reasoning capabilities
Cheapest powerful model (€0.08/€0.4 per Mtoken)

Best for Quick Iterations ⚡¶

Qwen3.5-9B - Near-instant responses

Fastest model in catalog
Still capable of complex tasks
Huge context window (262K)
Ultra-cheap (€0.1/€0.15)

Best for Code-Specific Tasks 🔧¶

Qwen3-Coder-30B-A3B-Instruct - Purpose-built for coding

Specialized code understanding
Fast inference (30B fp8)
Excellent code completion
Good pricing (€0.06/€0.22)

Best for Complex Reasoning 🧠¶

Qwen3.5-397B-A17B - Most powerful option

Handles complex architectural decisions
Multimodal + reasoning + function calling
Accept slower responses for maximum capability
Use for critical decisions, not routine edits

Backlinks¶

No other pages link to this page.