Llama 4 Maverick vs GLM 5
Llama 4 Maverick and GLM 5 sit at opposite ends of the price-vs-capability trade for late-2025/2026 releases. Maverick is Meta's MoE model (17B active params, 128 experts) with a 1,048,576-token context window and native image input. GLM 5 is Z.ai's newer, text-only agent-focused model with a 202,752-token window and reasoning support toggled on by default. Both are callable through OpenKey with one key and a flat 3% fee on provider list price.
Spec vs spec
| Spec | Llama 4 Maverick | GLM 5 |
|---|---|---|
| Context window | 1.0M | 203K |
| Max output | 16K | — |
| Input modalities | text, image | text |
| Output modalities | text | text |
| Knowledge cutoff | Aug 31, 2024 | — |
| Released | Apr 5, 2025 | Feb 11, 2026 |
| Reasoning | — | optional |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
meta-llama/llama-4-maverick
Input · 1M tokens
$0.150 + 3%$0.154
Output · 1M tokens
$0.600 + 3%$0.618
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.
meta-llama/llama-4-maverickCheaper
$2.78
$2.70 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.
| Llama 4 Maverick | GLM 5 | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 976 | #91 | 1307 | #15 |
| Code | 929 | #102 | 1295 | #16 |
| Data viz | 926 | #100 | 1269 | #22 |
| Game dev | 903 | #103 | 1299 | #15 |
| UI components | 955 | #94 | 1287 | #21 |
| Websites | 914 | #105 | 1290 | #18 |
Head-to-head preference voting. How we filter and rank
Pricing math on a real workload
Provider list price for Maverick is $0.15/M input and $0.60/M completion; on OpenKey that's $0.15 x 1.03 = $0.1545/M input and $0.60 x 1.03 = $0.618/M completion. GLM 5 lists at $0.60/M input and $1.92/M completion, working out to $0.618/M and $1.9776/M on OpenKey. Run 10M input tokens plus 2M output tokens and Maverick costs $2.70 total versus GLM 5's $9.84 — a 3.6x gap driven mostly by the 4x input-price ratio (0.25) and GLM 5's higher completion rate. GLM 5 does offer cache-read pricing at $0.12/M, which Maverick's provider doesn't list, so repeated-context workloads narrow the gap somewhat.
Coding and agent benchmarks
GLM 5 was built for long-horizon agent workflows and it shows in Design Arena: rank 3 in godotgamedev (elo 1237, 54.8% win rate), rank 6 in androidnative (elo 1244, 62% win rate), rank 10 in mobileapps, rank 13 in fullstack. Maverick has no entries in the agents arena at all — it's evaluated only in the general models arena, where it ranks 91-105 across 3d, codecategories, dataviz, gamedev, uicomponent, and website categories with win rates in the 33-41% range. On artificial_analysis metrics, Maverick posts a coding_index of 16.3 and agentic_index of just 1.3, consistent with it being a general-purpose multimodal model rather than an agent-tuned one.
Context and modality
Maverick's 1,048,576-token context is 5.17x larger than GLM 5's 202,752 tokens — the context_ratio the numbers confirm directly. If your task means ingesting a huge codebase, log dump, or document set in one shot, Maverick's window gives you more headroom. Maverick also accepts image input alongside text; GLM 5 is text-only. Trade-off: GLM 5's max_completion_tokens field is unset in its catalog record while Maverick caps completions at 16,384 tokens, so check your provider's actual output limit before assuming unlimited generation on GLM 5.
When to pick each
Maverick wins on raw cost, context length, and image input — good for high-volume, budget-sensitive pipelines or multimodal tasks. GLM 5 wins on agentic coding tasks (Android, full-stack, game dev, mobile) where its Design Arena agent-arena ranks are consistently in the top 15, and it ships with reasoning enabled by default (default_enabled: true, not mandatory), useful for tasks that benefit from a visible reasoning trace without forcing it on every call.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| High-volume batch text processing | Llama 4 Maverick | $2.70 vs $9.84 for the same 10M-in/2M-out workload |
| Full-stack or mobile app agent tasks | GLM 5 | Ranks 3-13 across godotgamedev, androidnative, mobileapps, fullstack agent categories |
| Multimodal input (images + text) | Llama 4 Maverick | Only model of the two with image input modality |
| Single-pass ingestion of massive documents | Llama 4 Maverick | 1,048,576-token context is 5.17x GLM 5's 202,752 |
| General website/game-dev generation | GLM 5 | Ranks 15-18 in gamedev, website, codecategories vs Maverick's 100-105 |
| Tasks needing a visible reasoning trace by default | GLM 5 | Ships with reasoning default_enabled: true |
Questions
- Which model is cheaper for large-scale text processing?
- Llama 4 Maverick, by a wide margin. On OpenKey, a 10M-input/2M-output workload costs $2.70 with Maverick versus $9.84 with GLM 5 — Maverick's input price is 4x lower ($0.1545/M vs $0.618/M after the 3% fee).
- Does either model support image input?
- Only Llama 4 Maverick. It takes text and image input and outputs text. GLM 5 is text-only on both input and output, per its modality field (text->text).
- Which one has a bigger context window?
- Llama 4 Maverick, at 1,048,576 tokens versus GLM 5's 202,752 — a 5.17x ratio. If your workload needs to fit a very large document or codebase in a single call, Maverick has the headroom.
- Which model is better for coding agents?
- GLM 5. It ranks 3rd in godotgamedev, 6th in androidnative, 10th in mobileapps, and 13th in fullstack on Design Arena's agents arena. Maverick has no agents-arena entries and posts an agentic_index of only 1.3 on artificial_analysis.