Mistral Large 3 2512 vs GLM 5
Mistral AIZ.aiboth via one key, provider price + 3%
Mistral Large 3 2512 (675B total params, 41B active, Apache 2.0) and GLM 5 (Z.ai's flagship for long-horizon agent workflows) both shipped within the last few months and target frontier-scale workloads. Mistral leans on mixture-of-experts efficiency and multimodal input; GLM 5 leans on agentic benchmarks and reasoning support. Both are available on OpenKey with one API key and a flat 3% fee on top of provider list price — no separate contracts to juggle.
Spec vs spec
| Spec | Mistral Large 3 2512 | GLM 5 |
|---|---|---|
| Context window | 262K | 203K |
| Input modalities | text, image, file | text |
| Output modalities | text | text |
| Released | Dec 1, 2025 | Feb 11, 2026 |
| Reasoning | — | optional |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
mistralai/mistral-large-2512
Input · 1M tokens
$0.500 + 3%$0.515
Output · 1M tokens
$1.50 + 3%$1.54
Cache read · 1M tokens
$0.050 + 3%$0.052
FEE — FLAT, EVERY MODEL3%
z-ai/glm-5
Input · 1M tokens
$0.600 + 3%$0.618
Output · 1M tokens
$1.92 + 3%$1.98
Cache read · 1M tokens
$0.120 + 3%$0.124
FEE — FLAT, EVERY MODEL3%
One workload, priced on both
10M input + 2M output tokens at each model's price, flat 3% fee included.
mistralai/mistral-large-2512Cheaper
$8.24
$8.00 provider + 3%
z-ai/glm-5
$10.14
$9.84 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| Mistral Large 3 2512 | GLM 5 | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1176 | #51 | 1307 | #15 |
| ASCII art | 1115 | #43 | 1192 | #26 |
| Code | 1191 | #56 | 1295 | #16 |
| Data viz | 1180 | #55 | 1269 | #22 |
| Game dev | 1146 | #65 | 1299 | #15 |
| SVG | 1050 | #62 | 1225 | #19 |
| UI components | 1157 | #59 | 1287 | #21 |
| Websites | 1205 | #53 | 1290 | #18 |
Head-to-head preference voting. How we filter and rank
Pricing math
Mistral Large 3 2512 costs $0.50/M input and $1.50/M output at provider rates; on OpenKey that's $0.515/M input and $1.545/M output (0.50 x 1.03, 1.50 x 1.03). GLM 5 costs $0.60/M input and $1.92/M output provider-side, or $0.618/M and $1.9776/M on OpenKey (0.60 x 1.03, 1.92 x 1.03). Mistral's input price is 0.83x GLM 5's — meaningfully cheaper on the input side. For a concrete workload of 10M input tokens plus 2M output tokens, Mistral Large 3 2512 costs $8.00 and GLM 5 costs $9.84. If you're running high-input-volume jobs (RAG, long documents, batch classification), that gap compounds fast.
Coding and agent benchmarks
GLM 5 is built for agent workflows and it shows: it has dedicated agent-arena benchmarks (androidnative rank 6, godotgamedev rank 3, mobileapps rank 10, fullstack rank 13) that Mistral Large 3 2512 has no equivalent entries for. On shared model-arena categories, GLM 5 outranks Mistral across the board — codecategories (1295 Elo, rank 16, 55.6% win rate) versus Mistral's 1191 Elo (rank 56, 47.6% win rate); website (1290 Elo, rank 18) versus Mistral's 1205 Elo (rank 53). Mistral also reports Artificial Analysis scores — intelligence index 15.9, coding index 20.1, agentic index 5.5 — but GLM 5 has no matching Artificial Analysis entry in this data, so that comparison stops at the Design Arena numbers, where GLM 5 leads everywhere.
Context and modality
Mistral Large 3 2512 supports a 262144-token context window versus GLM 5's 202752 tokens — a 1.29x ratio in Mistral's favor. That matters for long-document summarization or codebase-wide context. Modality is the bigger split: Mistral accepts text, image, and file input (text-only output), while GLM 5 is text-in, text-out only. If your pipeline needs to hand the model a screenshot, PDF, or scanned document, GLM 5 is disqualified regardless of its benchmark lead.
When to pick each
Pick GLM 5 when the job is agentic — multi-step coding, app scaffolding, tool-calling workflows — where its rank-3-to-rank-22 Design Arena results across gamedev, dataviz, uicomponent, and svg categories beat Mistral's rank-51-to-rank-65 results in the same categories. GLM 5 also natively supports `reasoning` as a parameter (default-enabled, not mandatory), useful if you want a togglable reasoning mode. Pick Mistral Large 3 2512 when input volume dominates cost, when you need multimodal input, or when you want the larger context window for long-document tasks. Mistral's Apache 2.0 license is also relevant if self-hosting or fine-tuning rights matter to your stack.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| Multi-step coding agent | GLM 5 | Ranks 3-16 across agent-arena categories (godotgamedev, fullstack) that Mistral has no entries for |
| High-volume RAG / document ingestion | Mistral Large 3 2512 | Input price is 0.83x GLM 5's rate, and the 10M-in/2M-out workload costs $8.00 vs $9.84 |
| Image or PDF input in the prompt | Mistral Large 3 2512 | Only Mistral accepts image and file input modalities; GLM 5 is text-only |
| Long-document context (200K+ tokens) | Mistral Large 3 2512 | 262144-token context window vs GLM 5's 202752 tokens |
| UI component / website generation | GLM 5 | Ranks 18-21 with 1287-1290 Elo vs Mistral's rank 53-59 with 1157-1205 Elo in the same categories |
| Self-hosting or fine-tuning rights | Mistral Large 3 2512 | Released under Apache 2.0 license |
Questions
- Which model is cheaper for a typical workload?
- Mistral Large 3 2512. On a 10M-input/2M-output workload it costs $8.00 total versus $9.84 for GLM 5 — driven mainly by Mistral's lower input price ($0.515/M vs $0.618/M on OpenKey after the 3% fee).
- Which model has a bigger context window?
- Mistral Large 3 2512, at 262144 tokens versus GLM 5's 202752 tokens — a 1.29x ratio. Useful if you're processing very long documents or large codebases in a single call.
- Does either model handle images?
- Only Mistral Large 3 2512. It accepts text, image, and file input modalities. GLM 5 is text-to-text only, so it's not an option for any workload that requires visual input.
- Which model wins on coding benchmarks?
- GLM 5. In the shared codecategories arena it scores 1295 Elo (rank 16, 55.6% win rate) versus Mistral's 1191 Elo (rank 56, 47.6% win rate) — a gap that holds across nearly every Design Arena category both models share.