Perplexity Quietly Added a Model 9x Cheaper: GLM 5.3 Flash at US$0.15 In, US$0.50 Out — When to Switch

Perplexity added a model to its official changelog in September with no press release, but the price gap is worth a look: GLM 5.3 Flash, at US$0.15 per million input tokens and US$0.50 per million output tokens.

In the same changelog, the full GLM 5.3 shipped in August at US$1.40 input and US$4.40 output. The Flash version is roughly one ninth the price on both.

Plain language: who is this for

This is not the Perplexity search you use in a browser. It is the Agent API and Router API — the interfaces your own code or automation calls. The selling point is one interface across many model providers: switching models means changing a name, not rewriting code.

So the practical meaning is: change perplexity/glm-5.3 to perplexity/glm-5.3-flash in the same program and the bill can drop to about a ninth — if the task is one a smaller model can handle.

The official numbers

(Both quoted directly from Perplexity’s changelog: the Flash entry from September 2026, the standard entry from August.)

In money you can feel: say your pipeline handles 2M input and 300K output tokens a day.

• GLM 5.3: 2 × 1.40 + 0.3 × 4.40 ≈ US$4.12/day
• GLM 5.3 Flash: 2 × 0.15 + 0.3 × 0.50 ≈ US$0.45/day

About US$110 a month apart. That is arithmetic, not a vendor claim about quality.

What to switch, what to leave alone

Nine times cheaper does not mean equally capable. The changelog publishes prices and model names only — there is no official capability comparison — so the following is operating guidance, not a benchmark result:

Reasonable to switch

• Classification, tagging, field extraction — fixed format, short answers
• High-volume repetitive batch work (sorting thousands of comments into three buckets)
• First-pass filtering: cheap model screens, expensive model handles only the hard cases

Do not switch blindly

• Multi-step reasoning that feeds a decision
• Long-form text customers will read
• Anything where an error costs human cleanup

The practical method: run both over the same real data and read 30 outputs by hand. Saving 9x is not a saving if 20% needs redoing.

Two changes in the same changelog that raise your bill

Easy to miss while looking at the new model:

1. The fast preset changed models. The Agent API fast preset now uses openai/gpt-5.6-luna with minimal reasoning effort and priority processing — and the changelog states priority processing costs 2× the model’s standard token prices. Dynamic preset users pick this up automatically; frozen configurations must be updated by hand.
2. Gemini 3.7 Flash got more expensive on August 27: US$0.75 input, US$0.075 cached input, US$3.75 output and reasoning tokens, per million.

One change goes your way: presets now use stable prompt cache keys automatically, so independent requests with the same preset reuse the shared prefix (system prompt and tool definitions). The changelog puts the saving at about 5% for preset-heavy applications, with no code change required.

Three things to do today

1. Open your usage bill and look at the share that is input tokens. Input-heavy pipelines gain the most, because that is where the gap is widest.
2. Check whether you are on the fast preset. If you are, know that you are being billed at the 2× priority rate.
3. Put the unchanging parts of your prompt (system prompt, tool definitions) first so caching hits — a hit takes input from US$0.15 down to US$0.03.

FAQ

Why can’t I find GLM 5.3 Flash on the pricing page?

Perplexity’s models and pricing pages are dynamically rendered and could not be read directly on 2026-09-05. Every price here comes from the wording of the official changelog entries. Evidence level: first-party, but a single official page rather than two cross-checked official documents.

Who makes GLM?

The GLM series is developed by Zhipu (Z.ai) in China. This article is about the price Perplexity charges to serve it, not a review of the model. Teams with data residency or compliance requirements should treat model provenance as part of the evaluation.

I use Perplexity in the browser — does this affect me?

Not directly. The web app and Pro/Max subscriptions are billed on a separate credit system, covered on our Perplexity tool page. The API is for people wiring models into their own software.

Sources: Perplexity official changelog at docs.perplexity.ai/changelog/changelog (September 2026 GLM 5.3 Flash entry; August GLM 5.3, preset prompt caching, fast preset, and August 27 Gemini 3.7 Flash pricing entries), read directly on 2026-09-05. Not available: capability comparison between GLM 5.3 and Flash, context lengths and rate limits — the official model pages are dynamic and could not be read that day, and we do not fill gaps with guesses. Currency: all figures are US dollars as published on the official English pages; no conversion applied. More: Perplexity tool page