GLM 5.3 Flash (batch)
Sold by 41 sellers, timed at 125 tok/s, scored 27.8 on the Intelligence Index, on 8 Design Arena boards, and with results on 2 openly licensed boards.
retrieved sha256 · 0bb53a77713f
What sellers charge
41 sellers · 5.64x from cheapest to dearestOne tick per seller, on a log scale of input price per million from $0.0399 at Relace to $0.225 at Fireworks. The $0.024 listed is the cheapest on offer; 41 sellers 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 69% 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
1,704,585 requests · 41 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
41 priced · 41 timed · cheapest first- Relacerelace/fp4$0.0399$0.0784$0.03fp4131k87 tok/s421.01s62,194 req99.02%
- Waferwafer/us$0.042$0.184$0.041nvfp4944knot matchednonenonenot matched100%
- OpenInferenceopen-inference/fp4$0.05$0.45$0.025fp4944k156 tok/s78680ms33,226 req100%
- Darkbloomdarkbloom/fp4$0.065$0.18$0.012fp429k25 tok/s103.08s429 req98.23%
- Particle AIparticle-ai/nvfp4/flex$0.065$0.325$0.02nvfp4131knot matchednonenonenot matched92.59%
- Waferwafer/nvfp4$0.065$0.24$0.064nvfp4944knot matchednonenonenot matched99.86%
- Tenstorrenttenstorrent/fp8$0.075$0.25$0.015fp8131k44 tok/s102.63s2,186 req100%
- Near AInear-ai/fp8$0.08$0.50$0.012fp8944k61 tok/s31614ms10,873 req100%
- Morphmorph/fp8$0.084$0.241$0.042fp8944k56 tok/s29270ms10,474 req99.93%
- Novitanovita/fp8 · 44% off$0.084$0.28$0.0168fp8131k60 tok/s292.29s15,327 req97.98%
- StreamLakestreamlake/fp8 · 41% off$0.0885$0.295$0.0177fp8128k41 tok/s242.41s15,459 req99.93%
- DigitalOceandigitalocean$0.0975$0.325$0.0195not declared131k117 tok/s67727ms146,387 req100%
- DekaLLMdekallm$0.10$0.50$0.03not declared944k64 tok/s34940ms18,303 req99.67%
- SiliconFlowsiliconflow/fp8$0.10$0.40$0.03fp8262k25 tok/s12954ms55 req99.85%
- DeepInfradeepinfra/fp4 · 33% off$0.1005$0.335$0.0201fp4944k109 tok/s54792ms25,149 req100%
- Coherecohere/nvfp4$0.105$0.34$0.02nvfp4131k112 tok/s63878ms155,705 req99.72%
- InferenceNetinference-net/fp4$0.108$0.35$0.035fp4262k94 tok/s50340ms25,384 req100%
- InferenceNetinference-net/us$0.108$0.35$0.035fp4262k84 tok/s53299ms16,931 req100%
- CoreWeavecoreweave/nvfp4$0.12$0.38$0.02nvfp4944k156 tok/s86549ms272,045 req100%
- Rekareka$0.12$0.30$0.03not declared944k80 tok/s44403ms10,127 req100%
- Riverriver$0.12$0.40$0.024not declared131k77 tok/s49709ms37,684 req96.04%
- Particle AIparticle-ai/nvfp4$0.13$0.65$0.04nvfp4131knot matchednonenonenot matched100%
- Makoramakora/fp4$0.14$0.24$0.03fp4131k51 tok/s171.7s9,045 req100%
- Baidubaidu/fp8$0.15$0.50$0.03fp8131k41 tok/s242.29s523 req100%
- BaseTenbaseten/fp8$0.15$0.50$0.03fp8131k124 tok/s66372ms88,354 req100%
- Crusoecrusoe/fp4$0.15$0.50$0.03fp4944k106 tok/s62811ms6,615 req100%
- Fireworksfireworks$0.15$0.50$0.03not declared944knot matchednonenonenot matched99.86%
- Friendlifriendli$0.15$0.50$0.03not declared944k108 tok/s74208ms17,184 req99.93%
- KTkt$0.15$0.50$0.03not declared944k140 tok/s351.47s3,899 req100%
- Mistralmistral$0.15$0.50$0.03not declared131k99 tok/s561.01s7,556 req100%
- Modalmodal/fp4$0.15$0.50$0.03fp4944k93 tok/s58394ms49,843 req100%
- Nebiusnebius/fp4$0.15$0.50$0.03fp4944k46 tok/s28771ms584 req98.39%
- Parasailparasail/fp8$0.15$0.50$0.03fp8944knot matchednonenonenot matched100%
- Togethertogether$0.15$0.50$0.03not declared944k167 tok/s104469ms253,107 req100%
- Venicevenice$0.15$0.50$0.03not declared131k120 tok/s56618ms9,106 req99.85%
- Z.AIz-ai/fp8$0.15$0.50$0.03fp8131k56 tok/s271.78s33,348 req100%
- Inceptroninceptron/fp8$0.16$0.50$0.04fp8944k101 tok/s53487ms14,400 req99.79%
- Sail Researchsail-research/fp4$0.175$0.625$0.0225fp4131k67.5 tok/s49774ms593 req100%
- Sail Researchsail-research/us$0.175$0.625$0.0225fp4131k65 tok/s32.82.16s58 req100%
- Parasailparasail/fast$0.1875$0.625$0.0375fp4944knot matchednonenonenot matched100%
- Fireworksfireworks/us$0.225$0.75$0.045not declared944knot matchednonenonenot matched98.26%
- Fireworksnot matchednot matchednot matchednot matchedno offer133 tok/s85883ms88,597 reqno offer
- Fireworksnot matchednot matchednot matchednot matchedno offer100 tok/s57519ms8,492 reqno offer
- Parasailnot matchednot matchednot matchednot matchedno offer189 tok/s95427ms11,674 reqno offer
- Parasailnot matchednot matchednot matchednot matchedno offer91 tok/s16.1674ms6,681 reqno offer
- Particle AInot matchednot matchednot matchednot matchedno offer148 tok/s75533ms102,544 reqno offer
- Particle AInot matchednot matchednot matchednot matchedno offer149 tok/s56687ms12,842 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer101 tok/s59356ms21,365 reqno offer
- Wafernot matchednot matchednot matchednot matchedno offer92 tok/s62355ms100,237 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: 33 of the 41 timed routes are slower than it, and the middle route runs at 93 tok/s. The widest median-to-p90 gap is Parasail's, 5.65x.
Fireworks publishes 2 offers and had 2 routes timed; Parasail publishes 2 offers and had 2 routes timed; Particle AI 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 . 35 of the 51 sellers with a reading fell under 99% on at least one of the 23 days held, 23 Oct 2026 to 21 Nov 2026: BaseTen (baseten/fp8/us) at 0.00% on 23 Oct 2026, Darkbloom at 56.41% on 2 Nov 2026, Relace (relace/fp4) at 57.57% on 3 Nov 2026, Nebius at 62.76% on 24 Oct 2026, and 31 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
21 parameters · 41 sellers- include_reasoningevery seller
- max_tokensevery seller
- reasoningevery seller
- reasoning_effortevery seller
- response_formatevery seller
- temperatureevery seller
- toolsevery seller
- top_pevery seller
- structured_outputs40 of 41not SiliconFlow
- tool_choice40 of 41not Tenstorrent
- stop37 of 41
- frequency_penalty36 of 41
- presence_penalty35 of 41
- top_k32 of 41
- seed29 of 41
- repetition_penalty27 of 41
- logprobs26 of 41
- top_logprobs26 of 41
- logit_bias19 of 41
- min_p15 of 41
- parallel_tool_calls2 of 41only Cohere and Inceptron
8 of these 21 are accepted by every one of the 41 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 $0.035 per million tokens, 3 in to 1 out: cheaper than 336 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.
- 3D scenes · Models1293 Elowin rate 55.0% · source rank 21the board, 113 ranked
- ASCII art · Models1257 Elowin rate 53.9% · source rank 14the board, 67 ranked
- Code · Models1265 Elowin rate 48.2% · source rank 30the board, 120 ranked
- Data visualisation · Models1257 Elowin rate 50.1% · source rank 33the board, 110 ranked
- Game development · Models1266 Elowin rate 46.4% · source rank 28the board, 120 ranked
- SVG · Models1269 Elowin rate 54.6% · source rank 18the board, 83 ranked
- UI components · Models1294 Elowin rate 51.9% · source rank 20the board, 116 ranked
- Websites · Models1256 Elowin rate 47.2% · source rank 40the board, 126 ranked
Position 54 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 · 29 results| Task | All runs | Score |
|---|---|---|
| Chess Puzzles | 167 | 0.21 |
| FrontierMath-Tier-4-v2-Private | 84 | 0.17 |
| FrontierMath-Tiers-1-3-v2-Private | 128 | 0.56 |
| GPQA diamond | 342 | 0.90 |
| Mystery Game Puzzles | 149 | 0.08 |
| OTIS Mock AIME 2024-2025 | 319 | 0.94 |
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 this run glm-5.3-flash_max. 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 | 97 |
| code_completion | 80.44 |
| code_generation | 77.47 |
| connections | 93.67 |
| consecutive_events | 84.30 |
| integrals_with_game | 45 |
| javascript | 63.64 |
| logic_with_navigation | 62 |
| math_comp | 93.14 |
| olympiad | 89.84 |
| paraphrase | 47.98 |
| plot_unscrambling | 60.26 |
| python | 60 |
| simplify | 46.95 |
| spatial | 96 |
| story_generation | 59.88 |
| summarize | 56.47 |
| tablejoin | 48.83 |
| tablereformat | 96.08 |
| theory_of_mind | 73.08 |
| typescript | 46.67 |
| typos | 78 |
| zebra_puzzle | 79.50 |
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-5.3-flash, 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 5.3 Flash (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.
