DeepSeek V4-Flash vs Qwen3.8 27B
DeepSeek V4-Flash scores higher on the intelligence index, 35 against 34. The cheapest machine here that runs DeepSeek V4-Flash is the Mac Studio M5 Ultra, 256GB, at $10,799. Qwen3.8 27B runs on the Framework Desktop, 32GB at $1,269, $9,530 less. On the Mac Studio M5 Ultra, 256GB, the cheapest machine here that runs both, they run at much the same speed: 49 and 48 tok/s, both estimated from memory bandwidth. At 500k tokens a day the Mac Studio M5 Ultra, 256GB pays for itself in 602 years running Qwen3.8 27B, against 1,736 years running DeepSeek V4-Flash.
| DeepSeek V4-Flash | Qwen3.8 27B | |
|---|---|---|
| Intelligence index | 35 | 34 |
| Class | Sonnet-class | Sonnet-class |
| Weights | 155 GB | 16 GB |
| Needs at 32k | 155 GB | 19 GB |
| Quantisation | UD-Q4_K_M | Q4_K_M |
| Parameters | 284B (13B active) | 27.8B |
| Max context | 1M | 256k |
| API price per 1M | $0.04732 in / $0.09464 out | $0.039 in / $1.25 out |
| Licence | MIT | Apache 2.0 |
| Machines here that run it | 1 of 37 | 30 of 37 |
| Cheapest machine that runs it | Mac Studio M5 Ultra, 256GB $10,799 | Strix Halo Framework Desktop, 32GB $1,269 |
Run DeepSeek V4-Flash on the Mac Studio M5 Ultra, 256GB · or Qwen3.8 27B on the Framework Desktop, 32GB
Jump to: Side by side on the Mac Studio M5 Ultra, 256GB · 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
The table above gives each model the cheapest machine that runs it, and those are two different machines, so nothing in it is a like-for-like race. The Mac Studio M5 Ultra, 256GB is the cheapest machine here that runs both, so this is the pair doing the same work on the same hardware.
| DeepSeek V4-Flash | Qwen3.8 27B | |
|---|---|---|
| Speed at 32k | 49 tok/s estimated | 48 tok/s estimated |
| Pay-back on this machine | Pays back in 1,736 years | Pays back in 602 years |
| API cost per month | $0.77 | $1.75 |
Run DeepSeek V4-Flash on the Mac Studio M5 Ultra, 256GB · or Qwen3.8 27B
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. Qwen3.8 27B 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 | DeepSeek V4-Flash | Qwen3.8 27B |
|---|---|---|
| 50ka few chats a day | 17,360 years | 6,021 years |
| 200klight assistant use | 4,340 years | 1,505 years |
| 1Ma moderate coding-assistant day | 868 years | 301 years |
| 4Mheavy coding with an agent | 217 years | 75 years |
| 20Magents running most of the day | 43 years | 15 years |
Run DeepSeek V4-Flash at 20M tokens a day · or Qwen3.8 27B
Machines that run one and not the other
Qwen3.8 27B needs 19 GB of memory at 32k of context and DeepSeek V4-Flash needs 155 GB. That puts Qwen3.8 27B on 29 of the 37 machines priced here that DeepSeek V4-Flash does not, starting at $1,269.
| Machine | Price | Memory | Speed on Qwen3.8 27B | Pay-back |
|---|---|---|---|---|
| Strix Halo Framework Desktop, 32GB | $1,269 | 32 GB | 10 tok/s estimated | Pays back in 91 years |
| Mac mini M6, 32GB | $1,299 | 32 GB | 6.9 tok/s estimated | Pays back in 82 years |
| AMD Radeon AI PRO R9700, 32GB | $1,299card only | 32 GB | 30 tok/s estimated | Pays back in 84 years |
| Strix Halo Corsair AI Workstation 300, 64GB | $1,700 | 64 GB | 10 tok/s estimated | Pays back in 121 years |
| Strix Halo Framework Desktop, 64GB | $1,959 | 64 GB | 10 tok/s estimated | Pays back in 140 years |
| NVIDIA GeForce RTX 5090, 32GB | $1,999card only | 32 GB | 66 tok/s estimated | Pays back in 123 years |
23 more, on the Qwen3.8 27B page.
DeepSeek V4-Flash is also head to head with Qwen3.8 Flash Next above it on the leaderboard. Two more models need much the same memory: MiniMax M2.7 and Inkling Small. Qwen3.8 27B is also head to head with Beam 1.0 501B-A23B below it on the leaderboard and Qwen3 32B, the last-generation Qwen nearest it in size. One more model needs much the same memory: Muse Glimmer 30B.
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. 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: every machine that runs DeepSeek V4-Flash · every machine that runs Qwen3.8 27B · every other match-up · both against the frontier · the quickest pay-back at each level of use