GLM 5.2 (batch)
Sold by 30 sellers, timed at 119 tok/s, scored 68.8 on the Coding Index, on 19 Design Arena boards, and with results on 2 openly licensed boards.
retrieved sha256 · 0bb53a77713f
What sellers charge
30 sellers · 330x from cheapest to dearestOne tick per seller, on a log scale of input price per million from $0.007 at Wafer to $2.31 at Alibaba. 5 sellers undercut the $0.322 listed; 25 are dearer. A faint tick declares no number format. Every seller is named on the endpoint ledger below.
The same sellers grouped by the number format each declares, on the same scale. The fp4 group alone spans 100% of the range, so the cheap end is not cheap for its number format. A declaration is checked by nobody: this groups the rows and certifies none.
Measured speed
79,989 requests · 30 endpoints · 30-minute windowOpenRouter's own timing of its routed traffic, weighted by requests, over one 30-minute window read . Every route is on the endpoint ledger below.
Every endpoint
30 priced · 30 timed · cheapest first- Waferwafer/us$0.007$6.80$0.006fp4944knot matchednonenonenot matched99.93%
- Baidubaidu/fp8 · 87% off$0.189$0.594$0.0351fp8131knot matchednonenonenot matched100%
- Waferwafer$0.19$8.00$0.18fp4944knot matchednonenonenot matched100%
- StreamLakestreamlake/fp8 · 80% off$0.28$0.88$0.052fp8128k38 tok/s291.7s11,618 req100%
- Relacerelace/fp4$0.31$2.40$0.075fp4131k90 tok/s51828ms5,030 req99.51%
- DeepInfradeepinfra/fp4/flex · 25% off$0.45$1.44$0.084fp4164knot matchednonenonenot matched100%
- DeepInfradeepinfra/fp4 · 25% off$0.5625$1.80$0.105fp4164knot matchednonenonenot matched99.57%
- Ambientambient/fp8$0.60$2.00$0.15fp8182k6 tok/s228s126 req92.15%
- CoreWeavecoreweave/fp4$0.62$2.42$0.14fp4944k218 tok/s95614ms18,816 req100%
- Morphmorph/fp8$0.643$1.978$0.217fp8944k89 tok/s61605ms2,336 req100%
- SiliconFlowsiliconflow/fp8 · 50% off$0.651$2.046$0.13fp8262k68 tok/s471.1s1,661 req100%
- DigitalOceandigitalocean$0.70$2.20$0.105not declared131k86.5 tok/s58.4567ms79 req100%
- Novitanovita/fp8 · 45% off$0.77$2.42$0.143fp8131k42 tok/s282.05s201 req100%
- DeepInfradeepinfra/fp4/priority · 25% off$0.84375$2.70$0.1575fp4164knot matchednonenonenot matched99.88%
- Alibabaalibaba/fp8$0.966$3.036$0.1932fp8131knot matchednonenonenot matched100%
- Alibabaalibaba/fp8/us-east-1$1.12$3.52$0.224fp8131knot matchednonenonenot matched100%
- BaseTenbaseten/fp8$1.40$4.40$0.14fp8262knot matchednonenonenot matched100%
- BaseTenbaseten/fp8/us$1.40$4.40$0.14fp8262knot matchednonenonenot matched100%
- Friendlifriendli$1.40$4.40$0.26not declared944k160 tok/s90.2509ms195 req99.44%
- Mistralmistral/zdr$1.40$4.40$0.14not declared131knot matchednonenonenot matched100%
- Nebiusnebius/fp4$1.40$4.40$0.14fp4944k90 tok/s57570ms2,548 req99.93%
- Parasailparasail/fp4$1.40$4.40$0.26fp4236k126 tok/s63628ms1,117 req100%
- Togethertogether$1.40$4.40$0.26not declared944k165 tok/s91196ms10,312 req100%
- Venicevenice/fp8$1.40$4.40$0.26fp8131k74 tok/s53.61.44s199 req100%
- Z.AIz-ai/fp8$1.40$4.40$0.26fp8131k48 tok/s31.81.94s1,762 req100%
- Mistralmistral/eu$1.54$4.84$0.154not declared128knot matchednonenonenot matched100%
- BaseTenbaseten/fast$2.10$6.60$0.21fp8262knot matchednonenonenot matched100%
- BaseTenbaseten/us/fast$2.10$6.60$0.21fp8262knot matchednonenonenot matched100%
- Baidubaidu/fast$2.25$7.88$0.56fp4131knot matchednonenonenot matched100%
- Alibabaalibaba/fast$2.31$7.26$0.462fp8131knot matchednonenonenot matched100%
- Alibabanot matchednot matchednot matchednot matchedno offer81 tok/s22.3666ms101 reqno offer
- Alibabanot matchednot matchednot matchednot matchedno offer59 tok/s34.91.55s100 reqno offer
- Alibabanot matchednot matchednot matchednot matchedno offer65 tok/s251.23s4 reqno offer
- Baidunot matchednot matchednot matchednot matchedno offer61 tok/s52958ms9,001 reqno offer
- Baidunot matchednot matchednot matchednot matchedno offer94 tok/s59.8811ms101 reqno offer
- BaseTennot matchednot matchednot matchednot matchedno offer187 tok/s87414ms562 reqno offer
- BaseTennot matchednot matchednot matchednot matchedno offer98 tok/s382.05s1,428 reqno offer
- BaseTennot matchednot matchednot matchednot matchedno offer80 tok/s361.25s346 reqno offer
- BaseTennot matchednot matchednot matchednot matchedno offer176 tok/s102439ms119 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer70.5 tok/s36884ms282 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer71 tok/s39.81.91s86 reqno offer
- DeepInfranot matchednot matchednot matchednot matchedno offer76 tok/s43701ms1,796 reqno offer
- Mistralnot matchednot matchednot matchednot matchedno offer105 tok/s52.5902ms200 reqno offer
- Mistralnot matchednot matchednot matchednot matchedno offer101 tok/s551.46s22 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer96 tok/s47417ms8,118 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer71 tok/s36416ms1,723 reqno offer
Each track runs from a route's p90, the rate nine in ten of its requests met or beat, up to its median. Weighted by requests, the figure above reports where the traffic went rather than what one route does: 24 of the 30 timed routes are slower than it, and the middle route runs at 83.8 tok/s. The widest median-to-p90 gap is Alibaba's, 3.63x.
Alibaba publishes 3 offers and had 3 routes timed; Baidu publishes 2 offers and had 2 routes timed; BaseTen publishes 4 offers and had 4 routes timed; DeepInfra publishes 3 offers and had 3 routes timed; Mistral publishes 2 offers and had 2 routes timed; Wafer publishes 2 offers and had 2 routes timed. The two documents share no route id, so those offers and routes are printed apart rather than paired.
Uptime is OpenRouter's own figure for each seller over the trailing day, read once a day, and the column prints the latest reading, taken . 30 of the 37 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: Nebius at 0.00% on 19 Nov 2026, Phala at 0.00% on 23 Oct 2026, Ambient at 54.50% on 15 Nov 2026, Baidu (baidu/fp8) at 76.55% on 11 Nov 2026, and 26 more. Source: OpenRouter's public endpoints API, one reading a day for each seller: 1,211 sellers carried one on the latest of the 23 days held.
Source: OpenRouter's public endpoints API. 1257 sellers across 359 models when the mirror ran, 186 of them served by more than one. List prices, not what anybody paid; a discount is the seller's own. Read . Speed is OpenRouter's own timing over one 30-minute window, read .
What each seller accepts
20 parameters · 30 sellers- include_reasoningevery seller
- max_tokensevery seller
- reasoningevery seller
- reasoning_effortevery seller
- temperatureevery seller
- toolsevery seller
- top_pevery seller
- response_format29 of 30not SiliconFlow
- tool_choice29 of 30not Ambient
- frequency_penalty27 of 30not DigitalOcean, StreamLake, and Z.AI
- stop27 of 30not DigitalOcean, SiliconFlow, and Z.AI
- presence_penalty26 of 30
- structured_outputs26 of 30
- top_k20 of 30
- seed18 of 30
- repetition_penalty16 of 30
- logprobs14 of 30
- top_logprobs14 of 30
- logit_bias9 of 30
- min_p8 of 30
7 of these 20 are accepted by every one of the 30 sellers that publish a list; the other 13 are declared by some sellers and not others, so what a request sending one of them gets depends on which seller serves it. Each list is the seller's own declaration in OpenRouter's endpoints API, read , and nothing here tested one.
Price position
342 priced modelsBlended $0.489 per million tokens, 3 in to 1 out: cheaper than 213 of the 342 priced models in the catalog. Each bar is one stretch of the catalog on a logarithmic price scale; the mark is this listing.
Measured quality
145 models with an index · 132 on a design arena boardEach mark is one board this model stands on, placed at the share of that board it scores above, and led to its number in the list under the rule. The columns and the ledger below carry the figures.
- Agentic game development · Agents1156 Elowin rate 49.7% · source rank 9the board, 25 ranked
- Android native · Agents1170 Elowin rate 52.8% · source rank 23the board, 41 ranked
- blender3d · Agents983 Elowin rate 36.1% · source rank 21the board, 25 ranked
- Full stack · Agents1215 Elowin rate 60.0% · source rank 15the board, 44 ranked
- Godot game development · Agents1142 Elowin rate 40.1% · source rank 19the board, 34 ranked
- HTML slides · Agents1072 Elowin rate 49.6% · source rank 15the board, 25 ranked
- Mobile apps · Agents1154 Elowin rate 47.6% · source rank 31the board, 49 ranked
- pdf · Agents1252 Elowin rate 59.2% · source rank 9the board, 36 ranked
- python-pptx slides · Agents1161 Elowin rate 48.4% · source rank 18the board, 29 ranked
- spreadsheet · Agents1013 Elowin rate 41.0% · source rank 25the board, 32 ranked
- Web apps · Agents1203 Elowin rate 55.6% · source rank 21the board, 47 ranked
- 3D scenes · Models1266 Elowin rate 52.7% · source rank 25the board, 113 ranked
- ASCII art · Models1174 Elowin rate 46.3% · source rank 33the board, 67 ranked
- Code · Models1275 Elowin rate 50.5% · source rank 27the board, 120 ranked
- Data visualisation · Models1281 Elowin rate 51.7% · source rank 17the board, 110 ranked
- Game development · Models1255 Elowin rate 49.0% · source rank 32the board, 120 ranked
- SVG · Models1195 Elowin rate 47.3% · source rank 32the board, 83 ranked
- UI components · Models1283 Elowin rate 55.2% · source rank 25the board, 116 ranked
- Websites · Models1276 Elowin rate 49.8% · source rank 26the board, 126 ranked
Position 56 of the 258 models Epoch has placed, which is not the population of either board above, on a scale anchored at Claude 3.5 Sonnet 130 and GPT-5 150. Epoch states that absolute values are meaningless by themselves and that comparisons between models are what the index is for, that values are not linearly related to benchmark accuracy, and that a highly specialised model may score low while being capable in its own domain.
Benchmark results
2 of 6 boards · 44 results| Task | All runs | Effort | Score |
|---|---|---|---|
| Chess Puzzles | 167 | max | 0.22 |
| low | 0.18 | ||
| none | 0.19 | ||
| EBR-bench | 34 | max | 0.10 |
| FrontierMath-Tier-4-v2-Private | 84 | max | 0.29 |
| FrontierMath-Tiers-1-3-v2-Private | 128 | max | 0.59 |
| low | 0.55 | ||
| none | 0.42 | ||
| GPQA diamond | 342 | max | 0.92 |
| low | 0.88 | ||
| none | 0.71 | ||
| MirrorCode | 48 | max | 0.37 |
| Mystery Game Puzzles | 149 | medium | 0.15 |
| low | 0.19 | ||
| none | 0.18 | ||
| minimal | 0.18 | ||
| OTIS Mock AIME 2024-2025 | 319 | max | 0.86 |
| low | 0.76 | ||
| none | 0.29 | ||
| SimpleQA Verified | 106 | max | 0.34 |
| SWE-Bench verified | 33 | max | 0.79 |
Mean accuracy, 0 to 1, on the task the row names, as Epoch AI runs it and averaged over repeated runs. All eighteen of their tasks, including the held-out FrontierMath tiers, named as they name them. The model id carries the effort setting the run used. Epoch AI Benchmarking Hub names these runs glm-5.2_max, glm-5.2_medium, glm-5.2_low, glm-5.2_none and glm-5.2_minimal, and the effort beside each score says which run it is. A run missing from a task's rows has no published result there, which is not a score of zero. All runs is how many runs the publisher scored on the task, every model's. Read from epoch.ai ↗, which dates its own data . Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
| Task | Score |
|---|---|
| AMPS_Hard | 98 |
| code_completion | 80.44 |
| code_generation | 78.87 |
| connections | 94 |
| consecutive_events | 79.28 |
| integrals_with_game | 76 |
| javascript | 63.64 |
| logic_with_navigation | 80 |
| math_comp | 96.08 |
| olympiad | 89.05 |
| paraphrase | 62.25 |
| plot_unscrambling | 60.73 |
| python | 55 |
| simplify | 55.67 |
| spatial | 94 |
| story_generation | 62.88 |
| summarize | 68.37 |
| tablejoin | 45.87 |
| tablereformat | 96.08 |
| theory_of_mind | 75 |
| typescript | 36.67 |
| typos | 74 |
| zebra_puzzle | 65.50 |
Percent correct on the task the row names, higher is better. All 23 published columns; there is no average to carry, because the site computes its headline in the browser from a weighting this does not reimplement. LiveBench names this run glm-5.2, with no effort setting written on it. Read from livebench.ai ↗, which dates its own data . LiveBench, Apache-2.0. Results produced by the LiveBench maintainers on questions refreshed monthly.
2 of the 6 openly licensed boards Vitna mirrors carry a result for GLM 5.2 (batch), each printed as its publisher published it and none added to another. MTEB, HELM, Vectara hallucination leaderboard and Aider polyglot have not published one under a name that matches its listing, and names are matched to the catalog exactly and never guessed. A task that is a link opens its column on /capability, with every run scored on it; an effort that is a link opens its run's row there, where the run is ranked on Vitna's scale. Every board, with its licence and what its publisher measures, is on /benchmarks.
Source: OpenRouter (openrouter.ai/models), as of 17 Nov 16:29 UTC. sha256 0bb53a77713f over 853,270 bytes.
Catalog: OpenRouter's public model list.
Intelligence, coding and agentic indexes: Artificial Analysis, as published by OpenRouter.
Elo, win rate and rank: Design Arena, as published by OpenRouter.
Source: each model's own Hugging Face repository (huggingface.co), as of 20 Nov 2026. 162 repositories read, 4 did not answer.
Licence, parameter count, model class and gating: each model's own Hugging Face repository, the one OpenRouter names for the listing.
A licence is the tag the uploader put on the repository, read from nothing else and verified by nobody.
A parameter count is the total the repository's safetensors files declare; for a mixture of experts that is every expert, not the share one token runs through.
Source: the Epoch AI Capabilities Index (epoch.ai/eci), as of 20 Nov 2026. 258 models placed over 2917 benchmark results.
Epoch AI Benchmarking Hub data, by Epoch AI, under CC BY 4.0.
Vitna ran none of these evaluations and adjusted none of these figures.
