ModelflareAuthenticity, speed and price analysis
About the site
Stable access to leading global models at a lower cost.
Modelflare is operated by US limited liability company Havenbyte LLC. It supports GPT, Claude, Gemini, Grok and common compatible protocols, plus over 40 image and video models.
Cache compensation
Selected OpenAI price-priority, stable and advanced groups offer daily cache-hit guarantees of 65%, 75% or 85%. If eligible requests fall short of the applicable target, Modelflare compensates the difference. Settled compensation can be viewed in the usage dashboard.
Top-up offers
A single US$20 top-up unlocks the GPT promotional group at a 0.015 pricing multiplier; a single US$50 top-up unlocks the Claude promotional group at a 0.03 multiplier.
Service details
- Minimum top-up5 USD
- Sign-up bonusUS$0.25
- Support responseAny time
- Payment methodsAlipay, WeChat Pay, Bank card, USDT, PayPal
- China invoiceSupported
- International invoiceSupported
- RefundsSupported
- PromotionsA single US$20 top-up unlocks the GPT promotional group at a 0.015 pricing multiplier; a single US$50 top-up unlocks the Claude promotional group at a 0.03 multiplier.
Strengths
- Established operation and a named entity: operating for over six months, under US limited liability company Havenbyte LLC.
- Broad model coverage: GPT, Claude, Gemini, Grok and other major models, plus over 40 image and video models.
- Distinctive cost controls: selected OpenAI groups offer cache-hit guarantees with compensation for eligible shortfalls, visible in the usage dashboard after settlement.
- Multiple group options: price-priority, stable, advanced and promotional groups cater to different needs.
- Low-barrier trial: US$0.25 in signup credits lets users try the service before topping up.
- Supports no-questions-asked refunds within 24 hours, reducing concerns about an initial top-up and trial.
- Business purchasing supported: invoices issued in China and corporate bank transfers make business payment and expense reimbursement easier.
Limitations
- Promotional groups require a top-up: a single US$20 payment unlocks GPT, while US$50 unlocks Claude.
- Cache compensation has eligibility conditions: it covers selected groups and qualifying requests, rather than every model and request.
- Group rules take time to understand: pricing multipliers, unlock requirements and compensation policies vary between groups.
(%)
(seconds)
coding session
availability
Comparison summary
GLM 5.3 Flash on Modelflare: a coding session costs ¥0.60, ranking 10 of 20 (median ¥0.80).
Authenticity
Model consistency
Output distribution for the same requests on this relay and on the official API · higher is better
Speed
Measured by time to first token (seconds) and output speed (tokens / second)
Time to first token
Seconds until the first token · lower is better
Price
Groups and models
| Group | Model | Availability | Input price | Output price | Pricing unit |
|---|---|---|---|---|---|
| glm-award | GLM 5.3 Flash | Not probed yet | ¥0.28 | ¥0.98 | Per million tokens |
| glm-stable | GLM 5.3 Flash | 97.91% | ¥0.40 | ¥1.40 | Per million tokens |
Cost per coding session
GLM 5.3 Flash · 2M input and 40k output tokens (¥) · split by token type · lower is better
Prices: cache hit, input and output
¥ / 1M tokens
Cache
Coding requests repeat much of their context, so the cache hit rate largely decides what you actually pay. Drag the hit rate and the costs and ranking below are reweighted with cache prices.
Cache-weighted session cost
GLM 5.3 Flash · of 2M input tokens, the hit share is billed at the cache price and the rest at the input price, plus 40k output (¥) · lower is better
Stability
Availability
Request success rate, last 24 hours · higher is better
Status of each model, last 24 hours
One cell per hour · green ok, yellow delayed, red failed
| Model | Group | Availability | Last 24 hours | First token | Samples |
|---|---|---|---|---|---|
| DeepSeek V4 Pro | deepseek-stable | 99.0% | 1.06 s | 194 | |
| Claude Opus 5 | claude-award | 99.0% | 1.21 s | 190 | |
| DeepSeek V4.1 Flash | deepseek-stable | 98.9% | 1.31 s | 189 | |
| GPT 6 Astra | openai-first-topup | 98.8% | 1.33 s | 165 | |
| Qwen 3.8 Max | qwen-stable | 98.8% | 1.06 s | 163 | |
| Gemini 3.8 Flash | gemini-stable | 98.7% | 1.40 s | 155 | |
| Grok 4.6 | grok-stable | 98.5% | 1.84 s | 135 | |
| Claude Sonnet 5 | claude-award | 98.5% | 1.77 s | 199 | |
| Claude Fable 5.1 | claude-first-topup | 98.5% | 1.34 s | 197 | |
| DeepSeek V4 Flash | deepseek-stable | 98.4% | 1.72 s | 188 | |
| Gemini 3.7 Flash | gemini-stable | 98.4% | 1.32 s | 186 | |
| GPT 5.6 Sol | openai-first-topup | 98.4% | 1.31 s | 123 | |
| Claude Opus 5.5 | claude-first-topup | 98.3% | 1.57 s | 179 | |
| GLM 5.3 | glm-stable | 98.3% | 0.96 s | 171 | |
| Grok 4.7 | grok-stable | 98.1% | 1.16 s | 161 | |
| GPT 6.1 Sol | openai-first-topup | 98.0% | 0.87 s | 151 | |
| GPT 5.5 | openai-first-topup | 97.8% | 1.01 s | 181 | |
| Claude Haiku 5.5 | claude-award | 97.7% | 1.79 s | 176 | |
| Gemini 3.6 Flash | gemini-stable | 97.6% | 1.79 s | 125 | |
| Gemini 3.1 Pro Preview | gemini-stable | 97.4% | 0.91 s | 154 | |
| GPT 6 Sol | openai-first-topup | 97.3% | 1.20 s | 151 | |
| Grok 4.5 | grok-stable | 97.3% | 1.42 s | 188 | |
| GPT 5.6 Terra | openai-first-topup | 97.2% | 1.04 s | 180 | |
| Kimi K3 | kimi-stable | 97.2% | 1.16 s | 176 | |
| Claude Fable 5 | claude-award | 97.0% | 0.94 s | 101 | |
| Gemini 3.5 Flash | gemini-stable | 97.0% | 1.78 s | 167 | |
| Claude Haiku 4.5 | claude-award | — | — | — | |
| Gemini 4 Argon | gemini-stable | — | — | — | |
| GLM 5.3 Flash | glm-award | — | — | — | |
| Claude Opus 4.8 | claude-award | — | — | — | |
| GLM 5.2 | glm-award | — | — | — |
FAQ
How much does GLM 5.3 Flash cost on Modelflare?
Input ¥0.28 and output ¥0.98 per 1M tokens; at 2M input and 40k output tokens, a coding session costs about ¥0.60, an API rate of 0.28.
How does the cache hit rate affect price?
Modelflare has not published a cache price for this route, so cache hits are billed at the input price.Measured hit rates will replace this assumption once available.
Which models does Modelflare support?
Folkbench currently lists 31: GPT 6 Astra, Claude Fable 5.1, Claude Opus 5.5, GPT 6.1 Sol, GPT 6 Sol, DeepSeek V4.1 Flash, Claude Haiku 5.5, Grok 4.7, GLM 5.3 Flash, Claude Fable 5, Claude Opus 5, Claude Opus 4.8, Claude Sonnet 5, GPT 5.6 Sol, Kimi K3, GLM 5.3, GPT 5.6 Terra, Gemini 4 Argon, Gemini 3.8 Flash, Grok 4.6, DeepSeek V4 Pro, Claude Haiku 4.5, Grok 4.5, Qwen 3.8 Max, GLM 5.2, Gemini 3.1 Pro Preview, Gemini 3.6 Flash, Gemini 3.7 Flash, Gemini 3.5 Flash, GPT 5.5, DeepSeek V4 Flash.
What does the data security level mean?
Folkbench rates relays from 0 to 4 based on public policies and verifiable evidence; Modelflare is currently level 4. The check covers leaks in responses and error messages only, and does not prove data retention or training policies.
How often is this data updated?
Availability and time to first token use a 24-hour window, and this page syncs live data every 15 minutes.