5 New LLMs Landed on TokenPAPA: Kimi K3.1 Pro, GLM-5.3-FlashX, MiniMax M3.1 Flash & Hy4
TokenPAPA added 5 new models in November 2026: Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, GLM-5.3-FlashX, MiniMax M3.1 Flash and Tencent Hy4. Pricing, context windows and use cases compared.
5 New LLMs Landed on TokenPAPA: Kimi K3.1 Pro, GLM-5.3-FlashX, MiniMax M3.1 Flash & Hy4
On 2026-11-05, TokenPAPA added five new large language models to its catalog: Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, GLM-5.3-FlashX, MiniMax M3.1 Flash and Hy4 (Tencent Hunyuan 4). Two of them are flagship-class agentic models from Moonshot and Tencent; three are fast, cheap workhorses from Zhipu and MiniMax. Every one of them is available right now through a single OpenAI-compatible API key at https://tokenpapa.ai/v1.
This post is the quickest way to decide which of the five belongs in your stack: the exact prices, what each model is actually good at, how the new models compare with the previous generation, and the code that switches you over in one line.
New TokenPAPA models in one paragraph: for agentic coding and long-context reasoning, Kimi K3.1 Pro (1M context, 3T-class, multimodal) at $2.10 / $10.60 per 1M input/output tokens; for the fastest experience, GLM-5.3-FlashX at up to 200 tokens/s and $0.32 / $1.10; for the cheapest high-throughput work, MiniMax M3.1 Flash at $0.16 / $0.63; for deep research with a 770B-class MoE and first-time vision, Hy4 at $1.00 / $2.80. All five share one key, so the switch is a one-line
model=change.
The five new models at a glance
These are the TokenPAPA platform rates per 1M tokens, taken directly from the pricing page on 2026-11-05. Cached input pricing matters more than ever because several of these models support long context reuse.
| Model ID | Vendor | Input / 1M | Output / 1M | Cached input / 1M | What it is |
|---|---|---|---|---|---|
kimi-k3.1-pro | Moonshot (Kimi) | $2.10 | $10.60 | $0.21 | 3T-class open-weight multimodal agentic model, 1M context |
kimi-k3.1-pro-highspeed | Moonshot (Kimi) | $4.10 | $21.30 | $0.21 | Faster companion of K3.1 Pro for low-latency use |
glm-5.3-flashx | Zhipu (GLM) | $0.32 | $1.10 | $0.09 | High-speed model, up to 200 tokens/s, 1M context |
minimax-m3.1-flash | MiniMax | $0.16 | $0.63 | $0.016 | Cheapest new model, high-throughput flash tier |
hy4 | Tencent (Hunyuan) | $1.00 | $2.80 | $0.05 | MoE flagship, 770B-class, first official vision support |
All five models accept OpenAI-style chat completions. Kimi K3.1 Pro and GLM-5.3-FlashX additionally accept image and video input, and Hy4 gains vision for the first time in the Hunyuan line.
Kimi K3.1 Pro: a 3T-class open-weight agentic model
Kimi K3.1 Pro is the first model in Moonshot's new K3.1 series, released on 2026-10-24. According to Moonshot, it is an open-weight, natively multimodal agentic model and the most capable 3T-class model the company has released, post-trained from Kimi K3 with substantially more reinforcement learning.
Key facts for developers:
- Context window: 1,048,576 tokens (1M) — the full class of "give it your whole codebase" workloads
- Modality: accepts text, images and video, generates text
- Family: 3T-class open-weight, available through the Kimi API platform and aggregators like TokenPAPA
- Price on TokenPAPA: $2.10 per 1M input, $10.60 per 1M output, $0.21 cached input
Key takeaway: Kimi K3.1 Pro is priced at roughly 33% below the previous kimi-k3 ($3.125 / $15.62) on TokenPAPA, which makes a 1M-context multimodal agentic model meaningfully cheaper per token than the generation it replaces.
Kimi K3.1 Pro Highspeed: the low-latency companion
Kimi K3.1 Pro Highspeed is the faster companion variant announced alongside K3.1 Pro. It targets interactive and latency-sensitive workloads where the extra output speed is worth a premium.
| Comparison | kimi-k3.1-pro | kimi-k3.1-pro-highspeed |
|---|---|---|
| Input / 1M tokens | $2.10 | $4.10 |
| Output / 1M tokens | $10.60 | $21.30 |
| Best for | Deep reasoning, agentic loops, multimodal | Interactive chat, fast code completion, latency-first APIs |
If your workload is dominated by long agentic loops where correctness matters more than wall-clock speed, the standard Pro is the better value. If you are building a chat product where users feel every second, Highspeed earns its premium.
GLM-5.3-FlashX: 200 tokens/s with 1M context
GLM-5.3-FlashX is Zhipu's high-speed model, supplied directly by Zhipu AI. The headline numbers, per Zhipu and Alibaba Cloud Model Studio documentation:
- Inference speed: up to 200 tokens/s
- Context: 1M tokens
- Input: natively accepts image, video and file input
- Price on TokenPAPA: $0.32 per 1M input, $1.10 per 1M output, $0.09 cached
This is the "flash but faster" tier: more expensive than the bare glm-5.3-flash ($0.12 / $0.42) but dramatically cheaper than glm-5.3 ($1.20 / $4.18), while keeping the long context and multimodal input of the newer line. For agents that stream a lot of tokens to the user, 200 tokens/s materially changes perceived quality.
MiniMax M3.1 Flash: the budget workhorse
MiniMax M3.1 Flash is the flash-tier entry of MiniMax's M3.1 line. Public technical detail is still thin right after launch, so here is what is verifiable from the TokenPAPA catalog: it is the cheapest of the five new models at $0.16 per 1M input and $0.63 per 1M output tokens, with cached input at just $0.016 — a 95% discount on cached reads.
At that price it sits below minimax-m3 ($0.328 / $1.30) and even below most previous-generation flash models. It is the natural default for high-volume extraction, classification, summarization and any pipeline where unit economics matter more than peak quality.
Hy4: Tencent's open-weight flagship gains vision
Hy4 is the official release of Tencent's next-generation Hunyuan model, following the hy4-preview open-source debut. Tencent says Hy4 significantly expands pretraining and post-training scale, introduces vision capability for the first time, and substantially improves instruction following, long-document understanding, long-horizon tasks and deep research.
Developer-relevant facts:
- Architecture: mixture-of-experts flagship; the preview measured 770B total parameters with 49B active per token, and the official release scales further
- Context: exceeding 1M tokens
- Open weight: released as an open-source model
- Price on TokenPAPA: $1.00 per 1M input, $2.80 per 1M output, $0.05 cached
Hy4 is the model to watch for research-heavy workloads: the combination of 1M+ context, native vision and open weights makes it a credible alternative to the big closed flagships at a fraction of their output price.
New vs old: how the prices moved
The clearest way to read this launch is against the previous generation, all from the same TokenPAPA pricing page:
| Model | Input / 1M | Output / 1M | Change |
|---|---|---|---|
kimi-k3 → kimi-k3.1-pro | $3.125 → $2.10 | $15.62 → $10.60 | Cheaper (~33% input) |
glm-5.3 → glm-5.3-flashx | $1.20 → $0.32 | $4.18 → $1.10 | Cheaper (~73% input) |
minimax-m3 → minimax-m3.1-flash | $0.328 → $0.16 | $1.30 → $0.63 | Cheaper (~51% input) |
hy3 → hy4 | $0.14 → $1.00 | $0.58 → $2.80 | More capable flagship, higher price |
Three of the four upgrades moved down in price while adding context or speed; Hy4 is the deliberate exception, a genuine flagship price point for a model that claims to be the strongest open-weight Hunyuan yet.
Which one should you use?
- Agentic coding and long-context reasoning →
kimi-k3.1-pro. 1M context, multimodal input, 3T-class quality, 33% cheaper than its predecessor. - Latency-sensitive products →
kimi-k3.1-pro-highspeedif you need flagship quality fast;glm-5.3-flashxif you want 200 tokens/s at a fraction of the price. - High-volume, cost-first pipelines →
minimax-m3.1-flash. $0.16 / $0.63 is the floor of this catalog. - Deep research, long documents, vision →
hy4. 770B-class MoE with 1M+ context and first-time vision. - Fastest general default →
glm-5.3-flashx. Speed, 1M context and multimodal input for less than a third of the flagship tier.
The practical point of an aggregator: you do not have to choose once. Wire up https://tokenpapa.ai/v1 with one key, and let each workload pick its own model per call.
How to call the new models
Every model on TokenPAPA uses the same OpenAI-compatible endpoint and the same key. Switching is one field:
from openai import OpenAI
client = OpenAI(base_url="https://tokenpapa.ai/v1", api_key="YOUR_KEY")
resp = client.chat.completions.create(
model="kimi-k3.1-pro", # or glm-5.3-flashx / minimax-m3.1-flash / hy4
messages=[{"role": "user", "content": "Summarize this in one paragraph."}],
)
print(resp.choices[0].message.content)The base URL is https://tokenpapa.ai/v1; the API key is created at https://tokenpapa.ai/keys. For the cheapest call, turn on prompt caching and you pay $0.21 (K3.1 Pro) or as little as $0.016 (M3.1 Flash) per 1M cached input tokens.
FAQ
Which new models were added to TokenPAPA in November 2026?
TokenPAPA added five new models on 2026-11-05: Kimi K3.1 Pro, Kimi K3.1 Pro Highspeed, GLM-5.3-FlashX, MiniMax M3.1 Flash and Tencent Hy4. All five share one API key and the same base URL at https://tokenpapa.ai/v1.
How much do the new TokenPAPA models cost?
Per 1M input and output tokens: Kimi K3.1 Pro $2.10 / $10.60; Kimi K3.1 Pro Highspeed $4.10 / $21.30; GLM-5.3-FlashX $0.32 / $1.10; MiniMax M3.1 Flash $0.16 / $0.63; Hy4 $1.00 / $2.80. Cached input tokens cost 80–95% less.
Which new model is best for high-throughput or low-latency workloads?
GLM-5.3-FlashX delivers up to 200 tokens/s and is the speed-focused default at $0.32 / $1.10. MiniMax M3.1 Flash is the cheapest high-throughput option at $0.16 / $0.63. Kimi K3.1 Pro Highspeed is the premium low-latency companion to the flagship.
Do I need a new API key to use the new models?
No. On TokenPAPA every model runs through one key and one OpenAI-compatible base URL, so switching models is a one-line change in your request — no new key, no new endpoint, no code rewrite.
Get started in minutes
All five models are live on TokenPAPA right now. Check the full catalog and per-model pricing at https://tokenpapa.ai/pricing, create your API key at https://tokenpapa.ai/keys, and start with the model that fits today's workload — you can switch to another one tomorrow with a single line of code.
Last updated 2026-11-06. Prices verified against the TokenPAPA pricing page on 2026-11-05.
How is this guide?