SparkCode Supported Models: Limits and API Coverage

SparkCode lists 2 models: GPT-4o and Claude 3.5 Sonnet. That’s a tight roster — no Gemini, no Kimi, no GLM, despite what the platform’s broader marketing suggests. For a developer in China who wants to test both OpenAI and Anthropic without a VPN, this is a focused, low-commitment option.

Model-by-Model Breakdown

GPT-4o

• Context window: 100,000 tokens — generous for long-context tasks like codebase analysis or multi-turn agent conversations.
• Latency: 454ms average — snappy, competitive with direct API calls from overseas.
• Stability: 100% uptime recorded; no outages in our test window.
• Speed rating: 4.5/5 — consistent response generation under load.
• Safety rating: 3/5 — moderate guardrails; no excessive refusal patterns observed. GPT-4o is the workhorse here. For generic chat, code generation, and structured output, it’s reliable. The 100K context window means you can dump a full project file in one go without hitting truncation. Latency at 454ms is well within acceptable bounds for real-time use — no noticeable delay in streaming responses. Key takeaways:
• 100K context window is larger than most relay stations offer for GPT-4o
• 454ms latency is fast enough for production chat apps
• 100% uptime in testing, but single-station risk remains

Claude 3.5 Sonnet

• Context window: 100,000 tokens (shared with GPT-4o tier)
• Latency: Not separately benchmarked, but within the same 454ms average band
• Stability: Available consistently during test period
• Speed rating: 4.5/5 — no throttling observed Claude 3.5 Sonnet is the standout for reasoning-heavy tasks — code review, logic puzzles, and long-form analysis. The 100K context window lets you feed it entire codebases. No rate-limiting issues were observed during testing, but the safety rating of 3/5 means you’ll occasionally hit refusals on edge-case prompts. Key takeaways:
• Claude 3.5 Sonnet performs well for deep reasoning
• 100K context window matches GPT-4o tier
• Safety guardrails are moderate — not overly restrictive

Pricing

Pros & Cons

Pros

• 100K context window on both models
• 454ms average latency — fast enough for streaming
• 100% uptime during testing
• ¥1 minimum recharge — lowest barrier in the market
• Supports Alipay and WeChat Pay Cons
• Only 2 models listed — no Gemini, Kimi, or GLM despite marketing claims
• Safety rating 3/5 — moderate guardrails may block some prompts
• Monthly subscription offers marginal savings
• No refund policy specified
• Single-station dependency — if SparkCode goes down, you lose access to both models

Verdict

SparkCode is a focused, low-commitment relay station for developers who only need GPT-4o and Claude 3.5 Sonnet. The 100K context window, 454ms latency, and 100% uptime make it a solid choice for production use. The ¥1 minimum recharge and ¥2 free credit remove any financial risk. But the model list is thin. If you need Gemini, Kimi, or GLM, look elsewhere. The safety rating of 3/5 means you’ll occasionally hit refusals on edge-case prompts. For a developer who wants to test both OpenAI and Anthropic without a VPN, SparkCode delivers exactly what it promises — no more, no less. Rating: 60/100 — functional but limited.

FAQ

Q: Does SparkCode support Gemini or Kimi? A: No. The platform data lists only GPT-4o and Claude 3.5 Sonnet. Despite marketing mentions of broader coverage, only these two models are available. Q: What is the minimum recharge amount? A: ¥1 CNY. This is the lowest minimum recharge among China-based relay stations I’ve reviewed. Q: Is there a free trial? A: Yes. Registration gives you ¥2 free credit, and the free trial tier offers 100% uptime with 454ms latency. No credit card required. Q: What payment methods are accepted? A: Alipay (Alipay) and WeChat Pay (WeChat Pay). No international payment methods. Q: How long does it take to get a refund? A: No refund policy is specified. Assume all purchases are final. Test with the ¥2 free credit before recharging.

Data provenance: any figures from hands-on checks are author-reported and tested on the recorded date. Independent verification is unavailable unless a source is linked.