Tencent Hy3 vs Ling 3.0 flash
Tencent Hy3 scores higher on the intelligence index, 26 against 25. None of the 37 machines here still sold at a published price runs Tencent Hy3 at 32k of context. Ling 3.0 flash runs on the Framework Desktop, 128GB, from $3,449. At 500k tokens a day that machine pays for itself in 1,352 years running Ling 3.0 flash. Shorten the context to 16k and one machine here holds both: the Mac Studio M5 Ultra, 256GB, at $10,799.
| Tencent Hy3 | Ling 3.0 flash | |
|---|---|---|
| Intelligence index | 26 | 25 |
| Class | Haiku-class | Haiku-class |
| Weights | 182 GB | 78 GB |
| Needs at 32k | 193 GB | 78 GB |
| Quantisation | Q4_K_M | Q4_K_M |
| Parameters | 295B (21B active) | 124B (5.1B active) |
| Max context | 256k | 256k |
| API price per 1M | $0.0825 in / $0.33 out | $0.021 in / $0.063 out |
| Licence | Apache 2.0 | MIT |
| Machines here that run it | 0 of 37 | 12 of 37 |
| Cheapest machine that runs it | none listed | Strix Halo Framework Desktop, 128GB $3,449 |
Run Tencent Hy3 on the Mac Studio M5 Ultra, 256GB at 16k of context · or Ling 3.0 flash on the Framework Desktop, 128GB
Jump to: Side by side on the Mac Studio M5 Ultra, 256GB at 16k of context · How much use it takes to pay for the machine · Machines that run one and not the other · The assumptions behind both columns
Side by side on the Mac Studio M5 Ultra, 256GB at 16k of context
Tencent Hy3 needs 193 GB at 32k of context. The biggest machine on this list is the Mac Studio M5 Ultra, 256GB, and it has 192 GB usable. Tencent Hy3's weights are 182 GB of that. The rest is the cache, and the cache is the part that shrinks when you ask for less context: at 16k the model needs 188 GB, which that machine holds. It already runs Ling 3.0 flash, so at 16k of context one machine here runs both. It costs $10,799. The rest of this page is what each model does with it.
| Tencent Hy3 | Ling 3.0 flash | |
|---|---|---|
| Speed at 16k | 20 tok/s estimated | 108 tok/s estimated |
| Pay-back on this machine | Pays back in 1,031 years | Pays back in 3,639 years |
| API cost per month | $1.49 | $0.36 |
Run Tencent Hy3 on the Mac Studio M5 Ultra, 256GB · or Ling 3.0 flash
How much use it takes to pay for the machine
Everything above is at 500k tokens a day. Pay-back moves with how much you actually run, so here are both models at the five levels of use the calculator names, on the Mac Studio M5 Ultra, 256GB at 16k of context. Tencent Hy3 pays for it sooner at every level of use, so which of them to run does not turn on how hard you work it.
| A day's use | Tencent Hy3 | Ling 3.0 flash |
|---|---|---|
| 50ka few chats a day | 10,305 years | 36,390 years |
| 200klight assistant use | 2,576 years | 9,097 years |
| 1Ma moderate coding-assistant day | 515 years | 1,819 years |
| 4Mheavy coding with an agent | 129 years | 455 years |
| 20Magents running most of the day | 26 years | 91 years |
Run Tencent Hy3 at 20M tokens a day · or Ling 3.0 flash
Machines that run one and not the other
Ling 3.0 flash needs 78 GB of memory at 32k of context and Tencent Hy3 needs 193 GB. That puts Ling 3.0 flash on 12 of the 37 machines priced here that Tencent Hy3 does not, starting at $3,449.
| Machine | Price | Memory | Speed on Ling 3.0 flash | Pay-back |
|---|---|---|---|---|
| Strix Halo Framework Desktop, 128GB | $3,449 | 128 GB | 41 tok/s estimated | Pays back in 1,352 years |
| Strix Halo GMKtec EVO-X2, 128GB | $3,500 | 128 GB | 41 tok/s estimated | Pays back in 1,505 years |
| Strix Halo GMKtec EVO-X3, 128GB | $3,600 | 128 GB | 41 tok/s estimated | Pays back in 1,548 years |
| Strix Halo Minisforum MS-S1 Max, 128GB | $3,799 | 128 GB | 41 tok/s estimated | Pays back in 1,489 years |
| Strix Halo Beelink GTR9 Pro, 128GB | $4,349 | 128 GB | 41 tok/s estimated | Pays back in 1,705 years |
| DGX Spark GB10 Grace Blackwell, 128GB | $4,699 | 128 GB | 51 tok/s estimated | Pays back in 1,717 years |
6 more, on the Ling 3.0 flash page.
Tencent Hy3 is also head to head with Inkling Small above it on the leaderboard. One more model needs much the same memory: GLM-5.3-Flash. Ling 3.0 flash is also head to head with MiniMax M2.7 below it on the leaderboard. Two more models need much the same memory: Qwen3.5 122B-A10B and Mistral Small 4 (119B-2603).
The assumptions behind both columns
Both columns use the same usage: 500k tokens a day at 15:1 input to output, 32k of context, $0.17 per kWh, and today's API prices held flat. The two sections that need one machine to hold both models are at 16k of context instead, which is the longest on the calculator's list where one does. Speeds marked estimated are worked out from memory bandwidth rather than measured, and pay-back scales with them. Where nobody rents an open model by the token, its API prices are the nearest hosted model's, named beside them. Machines are the 37 here with a published price that are still sold. Change any of it in the calculator.
More head to head: what Tencent Hy3 needs · every machine that runs Ling 3.0 flash · every other match-up · both against the frontier · the quickest pay-back at each level of use