GLM-5.3-Flash vs GLM-5.3
GLM-5.3-Flash and GLM-5.3 are within three points on the Models at Work leaderboard, which we treat as a tie (71 vs 68 out of 100). On independent tests alone, GLM-5.3 is ahead (66 vs 61). GLM-5.3-Flash is cheaper: 1,000 support replies cost about $0.65 at list price, against $5.96 on GLM-5.3, which is 9.2 times as much. On the Artificial Analysis Intelligence Index, one of the independent tests, GLM-5.3 leads 32 to 28.
| GLM-5.3-Flash · Z.ai | GLM-5.3 · Z.ai | |
|---|---|---|
| Leaderboard rank | #4 | #9 |
| Score (0–100) | 71 | 68 |
| Kind of work | Run it yourself | Run it yourself |
| Price per 1M tokens | $0.15 to read · $0.50 to write | $1.40 to read · $4.40 to write |
| Reads at once | 1M tokens (about 750,000 words) | 1M tokens (about 750,000 words) |
| Where you can get it | Z.ai API | Amazon Bedrock, Z.ai API |
| Can a company host it itself | Yes (open weights) | Yes (open weights) |
| Artificial Analysis Intelligence Index | 28 | 32 |
| Arena Text score | 1475 (rank 38) | 1478 (rank 27) |
| Arena WebDev score | 1609 (rank 26) | 1622 (rank 22) |
| Humanity's Sixth Sense (Scale) | 26.80 | Not measured |
| Scale SWE-Bench Pro V2 (mini-swe-agent) | Not measured | 84.30% |
| Scale MCP Atlas | Not measured | 84.20 |
| Humanity's Last Exam, Diamond (Scale) | Not measured | 16.40 |
| Best for | Bulk document and image processing, Routine steps planned by a stronger model, Running AI on your own servers | Running AI on your own servers, Automating multi-step back-office tasks, Cutting the bill on work now sent to a top-tier model |
Cost of 1,000 jobs
| 1,000 of these | GLM-5.3-Flash | GLM-5.3 |
|---|---|---|
| Support repliesthe customer message, the account notes and two help articles read; a 300-word reply written | $0.65 | $5.96 |
| Documents processeda two-page invoice or form read; the fields pulled out | $0.45 | $4.12 |
| Calls summariseda 40-minute call transcript read; a summary and next steps written | $1.80 | $17 |
| Contracts revieweda 40-page contract read against a checklist; a memo of issues written | $5.25 | $49 |
| Drafts writtena brief and one example read; a 600-word email or proposal section written | $0.63 | $5.62 |
Worked out from each vendor’s list price and a fixed size for each job. A product built on a model charges its own price on top. See what every model costs for the same jobs.
Choose GLM-5.3-Flash for
- Bulk document and image processing
- Routine steps planned by a stronger model
- Running AI on your own servers
Cheap enough for bulk work at roughly a tenth of GLM-5.3's list price, scoring 28 against 32 on Artificial Analysis's index; despite the name it is not fast.
Caveats: Slow on Z.ai's own service: about 60 tokens (word-pieces) a second, under half the speed of the fast models on this board. Users report less consistent output than the full GLM-5.3; the common pattern is a stronger model writing the plan and this one carrying it out. Running it yourself still needs a server with several high-end graphics cards. Not found on a major US cloud this run: Z.ai's own service or your own servers only.
Choose GLM-5.3 for
- Running AI on your own servers
- Automating multi-step back-office tasks
- Cutting the bill on work now sent to a top-tier model
The open model with the best user reports on the board, at about a quarter of Claude Opus 5.5's list price, and one a company can run on its own servers; read the licence first.
Caveats: The licence is unclear: Artificial Analysis lists a custom licence with commercial limits while Arena lists the permissive MIT licence. Have legal read it before you host it. It refuses fewer security-related requests than US models, which users report as a selling point; that is a governance question for everyone else. Z.ai is a Chinese vendor: its own service means data leaves for its servers. Amazon Bedrock lists the model at a higher price ($1.68 in / $5.28 out per million tokens).
What people running them say
GLM-5.3-Flash: A developer's month-long diary on Hacker News, a developer forum ('One month coding with GLM 5.3 Flash', 233 upvotes), reports a total bill of $68 for a month of having the model write and change software largely on its own; commenters agreed that figure 'is the headline'. The pattern they describe is to have the larger GLM-5.3 write the plan and Flash carry it out; one user notes that given an unambiguous plan 'GLM 5.3 Flash executes it just fine'. The complaints: on Z.ai's own service it is 'too slow for execution, despite the name', and the full model is 'way more consistent'.
GLM-5.3: One of the year's biggest launches on Hacker News, a developer forum (1,171 upvotes, 584 comments). People who ran it report that it works inside Claude Code, Anthropic's coding tool, with 'zero issues'; that work costing '$5 in tokens' on Claude costs '$0.50' on GLM; and that in code reviews it 'routinely finds bugs missed by Fable and Sol'. The controversy was security: users bragged of having it adapt attack code for operating systems that 'Claude and Opus outright refused', and others warned that the public download could be a weakened version.
Questions people ask
Which is better, GLM-5.3-Flash or GLM-5.3?
GLM-5.3-Flash and GLM-5.3 are within three points on the Models at Work leaderboard, which we treat as a tie (71 vs 68 out of 100). On independent tests alone, GLM-5.3 is ahead (66 vs 61). GLM-5.3-Flash is cheaper: 1,000 support replies cost about $0.65 at list price, against $5.96 on GLM-5.3, which is 9.2 times as much. On the Artificial Analysis Intelligence Index, one of the independent tests, GLM-5.3 leads 32 to 28.
When should I choose GLM-5.3-Flash over GLM-5.3?
Choose GLM-5.3-Flash for bulk document and image processing, routine steps planned by a stronger model, running ai on your own servers. Cheap enough for bulk work at roughly a tenth of GLM-5.3's list price, scoring 28 against 32 on Artificial Analysis's index; despite the name it is not fast.
When should I choose GLM-5.3 over GLM-5.3-Flash?
Choose GLM-5.3 for running ai on your own servers, automating multi-step back-office tasks, cutting the bill on work now sent to a top-tier model. The open model with the best user reports on the board, at about a quarter of Claude Opus 5.5's list price, and one a company can run on its own servers; read the licence first.
Which is cheaper, GLM-5.3-Flash or GLM-5.3?
GLM-5.3-Flash costs $0.15 per million tokens it reads (input) and $0.50 per million tokens it writes (output) on Z.ai's list price (Z.ai Pricing, read Nov 12, 2026). A token is about three-quarters of a word. GLM-5.3 costs $1.40 per million tokens it reads (input) and $4.40 per million tokens it writes (output) on Z.ai's list price (Z.ai Pricing, read Nov 12, 2026). A token is about three-quarters of a word.
What do 1,000 jobs cost on GLM-5.3-Flash and GLM-5.3?
At list price: 1,000 support replies cost $0.65 on GLM-5.3-Flash and $5.96 on GLM-5.3; 1,000 documents processed cost $0.45 on GLM-5.3-Flash and $4.12 on GLM-5.3; 1,000 calls summarised cost $1.80 on GLM-5.3-Flash and $17 on GLM-5.3; 1,000 contracts reviewed cost $5.25 on GLM-5.3-Flash and $49 on GLM-5.3; 1,000 drafts written cost $0.63 on GLM-5.3-Flash and $5.62 on GLM-5.3. A product built on either model charges its own price on top.
Scores come from the Models at Work leaderboard: Score is 0 to 100 and computed, not typed. 40% is independent testing (Artificial Analysis, Arena, Scale, Epoch, Vals and others; each test is scaled against the best score on this board, and tests of real work, such as finished freelance projects or professional tasks, count double the exams). 30% is price: the vendor's list price for a typical mix of reading and writing, on a fixed scale from $0.05 to $60 per million tokens, so a model that costs a fifth as much gains real ground. 15% is what people report after running it, from Signals and linked threads. 15% is fit: how much text it can take in at once and where a business can get it (the products and clouds that carry it), with a small credit for open weights. A missing component drops out and the rest are reweighted. A model with no independent test yet is provisional and ranks below every measured one. Within three points is a tie. Vendor-published figures are labelled and never counted. Prices are vendor list prices where published; a typical mix assumes three parts reading to one part writing.