Open-source showdown · 2026-08-22

GLM-5.2 vs DeepSeek V4 Pro
Picking a budget open flagship

Two open-source flagships with 1M context, both officially priced within a few dollars per million tokens. With DeepSeek's hike in effect and GLM's unchanged, the math needs redoing.

Updated 2026-11-08

Switch models on the same endpoint and measure the cost per token yourself; check /models for callable models.

#glm-5.2#deepseek-v4-pro#open-source flagship#comparison

Key points

$1.40 / $4.40

GLM-5.2 official API price

Per million tokens input/output. 744B-class MoE with 1M lossless context.

$0.66 / $1.98

V4 Pro official off-peak price

DeepSeek's official off-peak price (peak $1.32 / $3.96); the official pricing has been peak/off-peak since 08-17. QCode charges the same price around the clock — live rates on qcode.cc/models.

1M

Same-class context

GLM-5.2: 1M lossless; V4 Pro: 1M with 384K max output. Both swallow long repos whole.

87.9

V4 Pro's TB 2.1

DeepSeek's official figure (Minimal+max). GLM-5.2 is no longer Zhipu's strongest — that's 5.3.

Where each model stands now

GLM-5.2 (released in June) is Zhipu's previous-generation flagship: MIT open source, 1M lossless context, built for long-horizon tasks — note it's been succeeded by GLM-5.3 (08-14) as the flagship, but 5.2's price is unchanged and supply is steady. DeepSeek V4 Pro (0813 GA) is DeepSeek's current flagship: stronger agentic capability, an official 87.9 on TB 2.1, and peak/off-peak pricing since 08-17.

The price balance has shifted

Before the 08-17 hike, V4 Pro's price advantage was crushing; after it (peak output ¥27 per million), the board reshuffles — off-peak, DeepSeek is still the cheapest, while at peak hours GLM-5.2's flat $1.40/$4.40 is more competitive. The selection advice has changed from 'just take DeepSeek' to 'know your call-time distribution.'

Timeline

2026-06-17

GLM-5.2 released and open-sourced (MIT), 1M lossless context.

2026-08-13

DeepSeek V4 Pro GA; peak/off-peak repricing announced the same day.

2026-08-17

DeepSeek's new prices take effect; GLM-5.3 is out but 5.2's price is untouched.

Confirmed vs watch out

Confirmed

Both models' licenses, context and current prices have official sources; DeepSeek's peak/off-peak windows are in the official announcement. Both have real call volume on QCode.

Watch out

Don't mistake GLM-5.2 for Zhipu's strongest — that's 5.3 (coding +50%). 5.2's value is price and stability. And DeepSeek's '11x hike' is only the cache-hit input item; the main billing items rose far less.

Pick by call window

Off-peak / batch

DeepSeek V4 Pro: at half price off-peak it's the cheapest on the field — the default for nightly batches and evals.

Peak / steady budget

GLM-5.2: no peak swings, a flat $1.40/$4.40. Costs are more predictable for interactive daytime loads.

How to run your own numbers

First profile your call-time distribution: anything that can move to 00:30-08:30 Beijing time and weekends is unbeatable on DeepSeek's off-peak rate; interactive loads that must run at peak are steadier on GLM-5.2's flat pricing. Time-based routing only pays off when you call the two official APIs directly; on QCode both are switchable with one key at the same price around the clock, so switching by hour does not change the unit price.

On QCode

Both glm-5.2 and deepseek-v4-pro are on sale. QCode sets catalog prices independently, with the same price around the clock; deepseek-v4-pro is priced below glm-5.2, and live rates are on qcode.cc/models. A single key switches between them.

FAQ

Which is cheaper, GLM-5.2 or DeepSeek V4 Pro?

On the official APIs, DeepSeek V4 Pro is cheaper at every hour: $0.66 / $1.98 off-peak and $1.32 / $3.96 at peak, both below GLM-5.2's official $1.40 / $4.40. The hour only changes the size of the gap, and choosing by time of day only matters when you call the official APIs directly. On QCode the price is the same around the clock, and deepseek-v4-pro is likewise cheaper than glm-5.2; live rates are on qcode.cc/models.

Are both open-source?

Yes. GLM-5.2 is MIT-licensed, and V4 Pro's weights are also available on Hugging Face. Both are self-hostable.

Which is more capable?

V4 Pro has the newer agentic capability (official TB 2.1 of 87.9); GLM-5.2 has been overtaken by GLM-5.3 (coding +50%). For Zhipu's best pick 5.3; for cheap and steady pick 5.2.

How big are the context windows?

Both are 1M-token class. V4 Pro's max output is 384K; GLM-5.2's is 131K.

Is DeepSeek still worth it after the hike?

Off-peak it's still the cheapest open flagship. Workloads you can schedule are barely touched by the hike.

How do I route between them on QCode?

Switch model ids on the same key. QCode charges the same price around the clock, so there is no need to switch by time of day, and its deepseek-v4-pro rates are below glm-5.2 on both input and output (see /models) — choose by task.

Sources

Zhipu official (GLM-5.2 release, 06-17), DeepSeek's official repricing announcement (08-13) and V4 Pro GA announcement, QCode /models (scraped 08-22).

One key for both flagships — same price around the clock

glm-5.2 and deepseek-v4-pro are both callable on QCode with one key; the price is the same around the clock, and you switch by changing the model id.

Not sure where to start?

Try it for a week

Small Pack: $4 for $30 of quota, valid 7 days

Use it every day

Starter: $9.00/month with $5 of quota every day — just $0.060 per $1 of quota, vs about $0.113 for the Medium Pack