DeepSeek V4.1 Flash (1017) vs GLM-5.3
DeepSeek V4.1 Flash (1017) and GLM-5.3 are within three points on the Models at Work leaderboard, which we treat as a tie (72 vs 72 out of 100). On independent tests alone, GLM-5.3 is ahead (77 vs 64). DeepSeek V4.1 Flash (1017) is cheaper: 1,000 support replies cost about $1.38 at list price, against $5.96 on GLM-5.3, which is 4.3 times as much. On the Artificial Analysis Intelligence Index, one of the independent tests, DeepSeek V4.1 Flash (1017) leads 33 to 32.
| DeepSeek V4.1 Flash (1017) · DeepSeek | GLM-5.3 · Z.ai | |
|---|---|---|
| Leaderboard rank | #7 | #8 |
| Score (0–100) | 72 | 72 |
| Kind of work | High-volume work | Run it yourself |
| Price per 1M tokens | $0.30 to read · $1.20 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 | DeepSeek API, Gemini Enterprise Agent Platform (Google Cloud), Microsoft Foundry | Amazon Bedrock, Z.ai API |
| Can a company host it itself | Yes (open weights) | Yes (open weights) |
| Artificial Analysis Intelligence Index | 33 | 32 |
| Humanity's Last Exam, Diamond (Scale) | 20.50 | 16.40 |
| Humanity's Sixth Sense (Scale) | 30.10 | Not measured |
| Vals Finance Index (DeepSeek V4.1 Flash 1017) | 60.89% (rank 14) | Not measured |
| Vals Tax Agent Bench (DeepSeek V4.1 Flash 1017) | 44.35% (rank 8) | Not measured |
| Vals Legal Research Bench (DeepSeek V4.1 Flash 1017) | 45.19% (rank 23) | Not measured |
| Remote Labor Index (Scale) (listed as Deepseek V4.1 Flash) | 9.58% of projects automated | Not measured |
| Professional Reasoning Benchmark - Finance (Scale) | 47.15 | Not measured |
| Professional Reasoning Benchmark - Legal (Scale) | 47.26 | Not measured |
| Arena Text score | Not measured | 1478 (rank 27) |
| Scale SWE-Bench Pro V2 (mini-swe-agent) | Not measured | 84.30% |
| Scale MCP Atlas | Not measured | 84.20 |
| Arena WebDev score | Not measured | 1622 (rank 22) |
| Vals Finance Index | Not measured | 61.41% (rank 11) |
| Vals Tax Agent Bench | Not measured | 43.33% (rank 9) |
| Vals Legal Research Bench | Not measured | 49.04% (rank 12) |
| Best for | Bulk drafting and rewriting, First-pass replies at volume, Trials where data location is not a constraint | 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 | DeepSeek V4.1 Flash (1017) | GLM-5.3 |
|---|---|---|
| Support repliesthe customer message, the account notes and two help articles read; a 300-word reply written | $1.38 | $5.96 |
| Documents processeda two-page invoice or form read; the fields pulled out | $0.96 | $4.12 |
| Calls summariseda 40-minute call transcript read; a summary and next steps written | $3.72 | $17 |
| Contracts revieweda 40-page contract read against a checklist; a memo of issues written | $11 | $49 |
| Drafts writtena brief and one example read; a 600-word email or proposal section written | $1.41 | $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 DeepSeek V4.1 Flash (1017) for
- Bulk drafting and rewriting
- First-pass replies at volume
- Trials where data location is not a constraint
Close to the leaders on Vals AI's test of finance work (60.9% against 67.1%) at about a seventh of Kimi K3.1 Pro's list price, and sold by Google's and Microsoft's clouds as well as DeepSeek itself.
Caveats: DeepSeek's own service is in China and charges more at peak hours: $0.15 to $0.30 in and $0.60 to $1.20 out per million tokens. Read its data terms, or buy through a US cloud. Weak on the hardest work: 33 on Artificial Analysis's index against 60 or more for the leaders, and 9.6% of freelance projects finished to a client's standard on Scale's Remote Labor Index against 41.3% for Claude Fable 5.2. The weights are free to use commercially (MIT licence), per DeepSeek. DeepSeek's own test figures are not counted. Up from 18th to 7th because Vals AI's finance, tax and legal research tests now count: it gets 60.9%, 44.4% and 45.2% right. Places 6 to 8 share a score.
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
DeepSeek V4.1 Flash (1017): A 1,067-upvote thread on Hacker News, a developer forum, about the earlier Flash release said the low price is real and so is the gap to Opus at writing software; nothing newer was read this run, so this is unscored.
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, DeepSeek V4.1 Flash (1017) or GLM-5.3?
DeepSeek V4.1 Flash (1017) and GLM-5.3 are within three points on the Models at Work leaderboard, which we treat as a tie (72 vs 72 out of 100). On independent tests alone, GLM-5.3 is ahead (77 vs 64). DeepSeek V4.1 Flash (1017) is cheaper: 1,000 support replies cost about $1.38 at list price, against $5.96 on GLM-5.3, which is 4.3 times as much. On the Artificial Analysis Intelligence Index, one of the independent tests, DeepSeek V4.1 Flash (1017) leads 33 to 32.
When should I choose DeepSeek V4.1 Flash (1017) over GLM-5.3?
Choose DeepSeek V4.1 Flash (1017) for bulk drafting and rewriting, first-pass replies at volume, trials where data location is not a constraint. Close to the leaders on Vals AI's test of finance work (60.9% against 67.1%) at about a seventh of Kimi K3.1 Pro's list price, and sold by Google's and Microsoft's clouds as well as DeepSeek itself.
When should I choose GLM-5.3 over DeepSeek V4.1 Flash (1017)?
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, DeepSeek V4.1 Flash (1017) or GLM-5.3?
DeepSeek V4.1 Flash (1017) costs $0.30 per million tokens it reads (input) and $1.20 per million tokens it writes (output) on DeepSeek's list price (DeepSeek API pricing, read Nov 18, 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 18, 2026). A token is about three-quarters of a word.
What do 1,000 jobs cost on DeepSeek V4.1 Flash (1017) and GLM-5.3?
At list price: 1,000 support replies cost $1.38 on DeepSeek V4.1 Flash (1017) and $5.96 on GLM-5.3; 1,000 documents processed cost $0.96 on DeepSeek V4.1 Flash (1017) and $4.12 on GLM-5.3; 1,000 calls summarised cost $3.72 on DeepSeek V4.1 Flash (1017) and $17 on GLM-5.3; 1,000 contracts reviewed cost $11 on DeepSeek V4.1 Flash (1017) and $49 on GLM-5.3; 1,000 drafts written cost $1.41 on DeepSeek V4.1 Flash (1017) 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.