inthesingularity.ai
Tracker · GPU economics · priced through the day

GPU Crack Spreads

A B200 serving DeepSeek V4.1-Flash makes $23.37 of tokens an hour at the prices hosts are actually paid, and costs $7.87 an hour to rent: a crack of +$15.50. Like a refiner’s crack spread, what the machine makes less what it runs on, for seven GPUs and three open models, with yields from SemiAnalysis InferenceX.

H100†
−$2.44
$1.38 of tokens against $3.82 rent · owning +$0.21 · +$0.16 on the day
H200
−$2.88
$2.68 of tokens against $5.56 rent · owning +$1.46 · −$0.03 on the day
B200†
+$15.50
$23.37 of tokens against $7.87 rent · owning +$21.64 · +$1.02 on the day
B300†
+$69.82
$79.22 of tokens against $9.41 rent · owning +$76.96 · +$3.60 on the day
GB200†
+$71.07
$87.07 of tokens against $16.00 rent · owning +$85.21 · +$4.33 on the day
GB300†‡
+$175.51
$191.87 of tokens against $16.35 rent · owning +$189.56 · +$5.94 on the day
MI300X
−$3.20
$1.20 of tokens against $4.39 rent · owning +$0.25 · +$0.04 on the day

Crack per GPU-hour on DeepSeek V4.1-Flash, the most-used open model on OpenRouter: what the GPU’s benchmark output sells for at the prices hosts are paid, less its on-demand rent; “owning” sets the same tokens against SemiAnalysis’s cost of owning the GPU. Indicative, on token prices from Nov 20, 11:14 UTC. † InferenceX measured this GPU only at faster speeds than the target, so its yield, and crack, are if anything understated. ‡ Provisional rent: GB300 isn’t in the GPU index yet.

How the crack works

A refinery’s crack spread is what the products from a barrel of crude sell for, less the barrel. This tracker does the same for GPUs. The output of each GPU on each model comes from SemiAnalysis InferenceX, which benchmarks open models on every major GPU, continuously. Its agentic workload replays real coding-agent sessions, in which nearly all of each request’s input is already cached from earlier turns. We take the best configuration InferenceX has run for each GPU at the speed OpenRouter’s hosts actually deliver, 65 output tokens a second per user for V4.1-Flash. Each GPU’s yield is its input and output tokens a second and the share of input served from cache.

Those tokens are priced at what OpenRouter’s hosts are paid: each host’s prices, weighted by the requests it serves. The hour is priced at the GPU’s rent from our GPU Compute Price Index, on demand and, for commitments, on each provider’s own discount for the term. Token value less rent is the crack. A positive crack says renting the GPU and selling its tokens pays at the benchmark’s load. The break-even load says how busy the GPU must be to cover the rent. Power, networking and staff aren’t in it.

The board

Crack per GPU-hour against on-demand rent, with the load at which the GPU covers its rent. Where InferenceX has no result, the output a GPU would need to cover its rent, in output tokens a second, next to what a comparable GPU makes. Spot and commitment rents are in the curve below.

GPURentV4.1-FlashGLM-5.3Kimi K3
H100H100 SXM$3.82spot $2.49−$2.44†short of rent at full loadneeds 40 tok/sno InferenceX resultneeds 17 tok/sno InferenceX result
H200H200 SXM$5.56spot $3.51−$2.88short of rent at full loadneeds 58 tok/sB200 makes 298needs 25 tok/sB200 makes 73
B200B200 HGX$7.87spot $5.15+$15.50†covers rent at 34% load+$19.88†covers rent at 28% load+$8.10covers rent at 49% load
B300B300 HGX$9.41spot $6.35+$69.82†covers rent at 12% load+$15.51covers rent at 38% load+$5.28covers rent at 64% load
GB200GB200 NVL72$16.00spot $6.22 (1 provider) · 3 providers+$71.07†covers rent at 18% load+$36.08†covers rent at 31% load+$1.14covers rent at 93% load
GB300GB300 NVL72$16.35no spot price · provisional, 2 providers+$175.51†‡covers rent at 9% load+$71.84†‡covers rent at 19% load−$14.21†‡short of rent at full load
MI300XAMD MI300X$4.39spot $1.39 (1 provider)−$3.20short of rent at full loadneeds 46 tok/sno InferenceX resultneeds 20 tok/sno InferenceX result
H100rent $3.82 · spot $2.49
V4.1-Flash−$2.44†short of rent at full load
GLM-5.3needs 40 tok/sno InferenceX result
Kimi K3needs 17 tok/sno InferenceX result
H200rent $5.56 · spot $3.51
V4.1-Flash−$2.88short of rent at full load
GLM-5.3needs 58 tok/sB200 makes 298
Kimi K3needs 25 tok/sB200 makes 73
B200rent $7.87 · spot $5.15
V4.1-Flash+$15.50†covers rent at 34% load
GLM-5.3+$19.88†covers rent at 28% load
Kimi K3+$8.10covers rent at 49% load
B300rent $9.41 · spot $6.35
V4.1-Flash+$69.82†covers rent at 12% load
GLM-5.3+$15.51covers rent at 38% load
Kimi K3+$5.28covers rent at 64% load
GB200rent $16.00 · spot $6.22 (1 provider) · 3 providers
V4.1-Flash+$71.07†covers rent at 18% load
GLM-5.3+$36.08†covers rent at 31% load
Kimi K3+$1.14covers rent at 93% load
GB300rent $16.35 · no spot price · provisional, 2 providers
V4.1-Flash+$175.51†‡covers rent at 9% load
GLM-5.3+$71.84†‡covers rent at 19% load
Kimi K3−$14.21†‡short of rent at full load
MI300Xrent $4.39 · spot $1.39 (1 provider)
V4.1-Flash−$3.20short of rent at full load
GLM-5.3needs 46 tok/sno InferenceX result
Kimi K3needs 20 tok/sno InferenceX result

Per GPU-hour, US dollars. Rent: the GPU index, global, all providers, on demand, with spot below; GB300 isn’t in the index yet, so its rent is provisional (Oracle Cloud $18.00 on-demand, Verda $14.71 1-month commitment, the median of the two). † Measured only at faster speeds than the target: understated. Download every daily fixing: gpu-crack-daily.csv.

What an hour makes, and what it costs

Token value per GPU-hour at the prices hosts are paid, against on-demand rent and SemiAnalysis’s cost of owning the GPU, for every GPU and model with an InferenceX result. Past the rent line, renting pays; past the ownership line, owning does.

Token valueOn-demand rentCost of owningProvisional rent (GB300)
$0$50$100$150DeepSeek V4.1-FlashH100 on V4.1-Flash: $1.38 of tokens an hour (130 output tokens a second) against $3.82 on-demand rent, crack −$2.44; $1.17 an hour to own, owner's crack +$0.21H100−$2.44H200 on V4.1-Flash: $2.68 of tokens an hour (328 output tokens a second) against $5.56 on-demand rent, crack −$2.88; $1.22 an hour to own, owner's crack +$1.46H200−$2.88B200 on V4.1-Flash: $23.37 of tokens an hour (2,800 output tokens a second) against $7.87 on-demand rent, crack +$15.50; $1.73 an hour to own, owner's crack +$21.64B200+$15.50B300 on V4.1-Flash: $79.22 of tokens an hour (13,908 output tokens a second) against $9.41 on-demand rent, crack +$69.82; $2.26 an hour to own, owner's crack +$76.96B300+$69.82GB200 on V4.1-Flash: $87.07 of tokens an hour (13,354 output tokens a second) against $16.00 on-demand rent, crack +$71.07; $1.86 an hour to own, owner's crack +$85.21GB200+$71.07GB300 on V4.1-Flash: $191.87 of tokens an hour (15,785 output tokens a second) against $16.35 on-demand rent (provisional), crack +$175.51; $2.31 an hour to own, owner's crack +$189.56GB300+$175.51MI300X on V4.1-Flash: $1.20 of tokens an hour (145 output tokens a second) against $4.39 on-demand rent, crack −$3.20; $0.95 an hour to own, owner's crack +$0.25MI300X−$3.20GLM-5.3B200 on GLM-5.3: $27.75 of tokens an hour (298 output tokens a second) against $7.87 on-demand rent, crack +$19.88; $1.73 an hour to own, owner's crack +$26.02B200+$19.88B300 on GLM-5.3: $24.92 of tokens an hour (240 output tokens a second) against $9.41 on-demand rent, crack +$15.51; $2.26 an hour to own, owner's crack +$22.66B300+$15.51GB200 on GLM-5.3: $52.08 of tokens an hour (768 output tokens a second) against $16.00 on-demand rent, crack +$36.08; $1.86 an hour to own, owner's crack +$50.22GB200+$36.08GB300 on GLM-5.3: $88.19 of tokens an hour (1,145 output tokens a second) against $16.35 on-demand rent (provisional), crack +$71.84; $2.31 an hour to own, owner's crack +$85.88GB300+$71.84Kimi K3B200 on Kimi K3: $15.97 of tokens an hour (73 output tokens a second) against $7.87 on-demand rent, crack +$8.10; $1.73 an hour to own, owner's crack +$14.24B200+$8.10B300 on Kimi K3: $14.69 of tokens an hour (63 output tokens a second) against $9.41 on-demand rent, crack +$5.28; $2.26 an hour to own, owner's crack +$12.43B300+$5.28GB200 on Kimi K3: $17.14 of tokens an hour (76 output tokens a second) against $16.00 on-demand rent, crack +$1.14; $1.86 an hour to own, owner's crack +$15.28GB200+$1.14GB300 on Kimi K3: $2.15 of tokens an hour (10 output tokens a second) against $16.35 on-demand rent (provisional), crack −$14.21; $2.31 an hour to own, owner's crack −$0.16GB300−$14.21
$0$50$100$150DeepSeek V4.1-FlashH100 on V4.1-Flash: $1.38 of tokens an hour (130 output tokens a second) against $3.82 on-demand rent, crack −$2.44; $1.17 an hour to own, owner's crack +$0.21H100−$2.44H200 on V4.1-Flash: $2.68 of tokens an hour (328 output tokens a second) against $5.56 on-demand rent, crack −$2.88; $1.22 an hour to own, owner's crack +$1.46H200−$2.88B200 on V4.1-Flash: $23.37 of tokens an hour (2,800 output tokens a second) against $7.87 on-demand rent, crack +$15.50; $1.73 an hour to own, owner's crack +$21.64B200+$15.50B300 on V4.1-Flash: $79.22 of tokens an hour (13,908 output tokens a second) against $9.41 on-demand rent, crack +$69.82; $2.26 an hour to own, owner's crack +$76.96B300+$69.82GB200 on V4.1-Flash: $87.07 of tokens an hour (13,354 output tokens a second) against $16.00 on-demand rent, crack +$71.07; $1.86 an hour to own, owner's crack +$85.21GB200+$71.07GB300 on V4.1-Flash: $191.87 of tokens an hour (15,785 output tokens a second) against $16.35 on-demand rent (provisional), crack +$175.51; $2.31 an hour to own, owner's crack +$189.56GB300+$175.51MI300X on V4.1-Flash: $1.20 of tokens an hour (145 output tokens a second) against $4.39 on-demand rent, crack −$3.20; $0.95 an hour to own, owner's crack +$0.25MI300X−$3.20GLM-5.3B200 on GLM-5.3: $27.75 of tokens an hour (298 output tokens a second) against $7.87 on-demand rent, crack +$19.88; $1.73 an hour to own, owner's crack +$26.02B200+$19.88B300 on GLM-5.3: $24.92 of tokens an hour (240 output tokens a second) against $9.41 on-demand rent, crack +$15.51; $2.26 an hour to own, owner's crack +$22.66B300+$15.51GB200 on GLM-5.3: $52.08 of tokens an hour (768 output tokens a second) against $16.00 on-demand rent, crack +$36.08; $1.86 an hour to own, owner's crack +$50.22GB200+$36.08GB300 on GLM-5.3: $88.19 of tokens an hour (1,145 output tokens a second) against $16.35 on-demand rent (provisional), crack +$71.84; $2.31 an hour to own, owner's crack +$85.88GB300+$71.84Kimi K3B200 on Kimi K3: $15.97 of tokens an hour (73 output tokens a second) against $7.87 on-demand rent, crack +$8.10; $1.73 an hour to own, owner's crack +$14.24B200+$8.10B300 on Kimi K3: $14.69 of tokens an hour (63 output tokens a second) against $9.41 on-demand rent, crack +$5.28; $2.26 an hour to own, owner's crack +$12.43B300+$5.28GB200 on Kimi K3: $17.14 of tokens an hour (76 output tokens a second) against $16.00 on-demand rent, crack +$1.14; $1.86 an hour to own, owner's crack +$15.28GB200+$1.14GB300 on Kimi K3: $2.15 of tokens an hour (10 output tokens a second) against $16.35 on-demand rent (provisional), crack −$14.21; $2.31 an hour to own, owner's crack −$0.16GB300−$14.21

The crack across the rent curve

DeepSeek V4.1-Flash: the crack per GPU-hour against each way of renting the GPU, against the cost of owning it, and the most a host could pay for it at 50% load. Commitment rents are on-demand rent times each provider’s own discount for the term, the median across providers; * where too few providers quote the term for this GPU, the discount across all GPUs.

GPUSpotOn-demand1 month1 year3 yearsOwningPays at 50% load
H100†$1.38 of tokens−$1.11on $2.49−$2.44on $3.82−$1.84on $3.22−$1.26on $2.64−$0.30on $1.68+$0.21on $1.17$0.69short of a 1-year lease
H200$2.68 of tokens−$0.83on $3.51−$2.88on $5.56−$2.04on $4.72−$1.15on $3.83+$0.11on $2.57+$1.46on $1.22$1.34short of a 1-year lease
B200†$23.37 of tokens+$18.23on $5.15+$15.50on $7.87+$15.51on $7.87+$17.75on $5.62*+$19.87on $3.50*+$21.64on $1.73$11.69covers a 1-year lease
B300†$79.22 of tokens+$72.88on $6.35+$69.82on $9.41+$70.01on $9.22*+$72.51on $6.72*+$75.04on $4.18*+$76.96on $2.26$39.61covers a 1-year lease
GB200†$87.07 of tokens+$80.85on $6.22+$71.07on $16.00+$71.39on $15.68*+$75.65on $11.42*+$79.96on $7.11*+$85.21on $1.86$43.54covers a 1-year lease
GB300†‡$191.87 of tokens—+$175.51on $16.35+$175.84on $16.03*+$180.19on $11.68*+$184.60on $7.27*+$189.56on $2.31$95.93covers a 1-year lease
MI300X$1.20 of tokens−$0.19on $1.39−$3.20on $4.39−$3.11on $4.31*−$1.94on $3.14*−$0.76on $1.95*+$0.25on $0.95$0.60short of a 1-year lease

Cost of owning: SemiAnalysis’s total cost of ownership per GPU-hour, as InferenceX publishes it. The last column is token value at 50% load, the rent a host could pay before running at a loss.

What the cracks say

Hosts are paid less than they post. Weighted by the requests each serves, V4.1-Flash sells for 10% more than the median host’s posted prices; the busiest host, DeepInfra, carries 15% of its requests.

On V4.1-Flash, B200, B300, GB200 and GB300 cover their rent; H100, H200 and MI300X don’t, even at full load at the speed hosts serve: an H100 makes $1.38 of tokens an hour against $3.82 of rent. Owning changes that: at SemiAnalysis’s cost of ownership the H100, H200 and MI300X earn +$0.21, +$1.46 and +$0.25 an hour. On those GPUs the lease takes the margin.

If renters compete the on-demand crack away, the break-even load is the utilization the market is pricing in: 34% for a B200, 12% for a B300 and 18% for a GB200 on V4.1-Flash.

Daily crack fixings

Crack per GPU-hour against on-demand rent at 16:00 UTC each day, from the first fixing on October 28. Each series follows a class of model, not one model, and moves to a successor by rule (marked). Each panel has its own scale.

H100H200B200B300GB200GB300MI300X
Efficient: DeepSeek V4.1-Flashcrack per GPU-hour
$0$50$100Oct 28Nov 4Nov 11Nov 18GB300 +$113.21GB200 +$77.78B300 +$75.46B200 +$17.20H100 −$2.33H200 −$2.70MI300X −$3.52
$0$50$100Oct 28Nov 4Nov 11Nov 18GB300 +$113.21GB200 +$77.78B300 +$75.46B200 +$17.20H100 −$2.33H200 −$2.70MI300X −$3.52
Large: GLM-5.3crack per GPU-hour
$0$20$40$60Oct 28Nov 4Nov 11Nov 18GB300 +$47.67GB200 +$22.61B200 +$12.61B300 +$8.70
$0$20$40$60Oct 28Nov 4Nov 11Nov 18GB300 +$47.67GB200 +$22.61B200 +$12.61B300 +$8.70
Frontier-size: Kimi K3crack per GPU-hour
−$10$0$10$20$30Oct 28Nov 4Nov 11Nov 18B200 +$6.31B300 +$3.11GB200 −$0.69GB300 −$14.26
$0$20Oct 28Nov 4Nov 11Nov 18B200 +$6.31B300 +$3.11GB200 −$0.69GB300 −$14.26

What moved the V4.1-Flash crack since the previous fixing

GPUCrackToken pricesYieldRentOver 7 days
H100since Nov 18+$0.16+$0.06$0.00+$0.09+$0.14
H200since Nov 18−$0.03+$0.13$0.00−$0.16+$0.28
B200since Nov 18+$1.02+$1.11$0.00−$0.10+$2.09
B300since Nov 18+$3.60+$4.11$0.00−$0.51+$53.01
GB200since Nov 18+$4.33+$4.33$0.00$0.00+$71.92
GB300since Nov 18+$5.94+$5.94$0.00$0.00+$11.81
MI300Xsince Nov 18+$0.04+$0.04$0.00$0.00+$0.08

Yields over time

Output tokens a second per GPU from InferenceX’s best configuration at each model’s target speed, at every InferenceX run, log scale. The same GPU makes more as engines improve; the crack picks that up on the day InferenceX publishes it.

H100H200B200B300GB200GB300MI300X
DeepSeek V4.1-Flashoutput tok/s per GPU at 65 tok/s per user
1002005001K2K5K10K20KOct 30Nov 6Nov 13Nov 20GB300 15,785B300 13,908GB200 13,354B200 2,800H200 328MI300X 145H100 130
1005001K5K10KOct 30Nov 6Nov 13Nov 20GB300 15,785B300 13,908GB200 13,354B200 2,800H200 328MI300X 145H100 130
GLM-5.3output tok/s per GPU at 45 tok/s per user
2005001KOct 30Nov 6Nov 13GB300 1,145GB200 768B200 298B300 240
2005001KOct 30Nov 6Nov 13GB300 1,145GB200 768B200 298B300 240
Kimi K3output tok/s per GPU at 25 tok/s per user
100200Oct 27Oct 29Oct 31Nov 2Nov 4GB300 206GB200 76B200 73B300 63
100200Oct 27Oct 29Oct 31Nov 2Nov 4GB300 206GB200 76B200 73B300 63

Token prices

Per million tokens. Realized: each host’s prices weighted by the requests it served in the 24 hours to the print, which is what the crack uses. Posted: the median host’s list, each host counted once at its cheapest standard endpoint. Speed: the p90 speed of hosts’ requests (90% reach it or beat it), weighted the same way, beside the target the yields are taken at.

ModelHostsInputCached inputOutputPosted (median host)First partySpeed
V4.1-Flash4.09M requests in 30 min43$0.139$0.00564$0.648$0.15 / $0.0036 / $0.65110% of posted, for this traffic$0.15 / $0.003 / $0.60DeepSeek off-peak72 tok/starget 65
GLM-5.3230K requests in 30 min36$0.636$0.144$2.80$1.05 / $0.155 / $4.1781% of posted, for this traffic$1.40 / $0.26 / $4.40Z.ai list56 tok/starget 45
Kimi K333K requests in 30 min20$1.77$0.252$14.27$2.92 / $0.30 / $15.0079% of posted, for this traffic$3.00 / $0.30 / $15.00Moonshot list32 tok/starget 25
GLM-5.3-Flashprices only: no InferenceX result35$0.111$0.0275$0.382$0.12 / $0.03 / $0.5090% of posted, for this traffic$0.15 / $0.03 / $0.50Z.ai list53 tok/s
Every V4.1-Flash host (43) · 1 left out as outliers
HostShare of requestsInputCached inputOutputDiscountSpeed
DeepInfrafp815%$0.098$0.00294$0.29451%67
InferenceNetfp812%$0.08$0.008$0.666—88
Wafermxfp412%$0.036$0.01$0.50—78
Togetherunknown7%$0.30$0.006$1.20—114
CoreWeavefp85%$0.13$0.002$0.65—78
Pearl Researchfp84%$0.05$0.003$0.30—48
Nebiusfp44%$0.18$0.0036$0.7240%87
Coherefp84%$0.115$0.0025$0.40—64
Morphfp84%$0.059$0.005$0.52—70
Relacefp44%$0.12$0.003$0.50—72
Fireworksunknown3%$0.15$0.003$0.60—60
Alibabaunknown3%$0.18$0.0036$0.72—64
PrimeIntellectunknown3%$0.0437$0.0024$0.70—66
DigitalOceanunknown2%$0.18$0.0036$0.72—78
Baidufp82%$0.30$0.006$1.20—28
OpenInferencefp42%$0.06$0.002$0.80—78
Ionstreamunknown2%$0.21$0.0105$0.62—84
Riverunknown2%$0.12$0.0024$0.48—60
StepFunfp82%$0.105$0.0021$0.4265%51
Particle AIunknown1%$0.06$0.006$0.50—63
Novitafp81%$0.15$0.003$0.6050%81
Parasailfp81%$0.30$0.006$1.20—91
Rekaunknown1%$0.12$0.008$0.6420%60
BaseTenunknown1%$0.30$0.007$1.20—79
Crusoefp81%$0.29$0.007$1.20—71
ModelRunfp41%$0.17$0.006$0.67—76
Cloudflareunknown1%$0.10$0.003$0.60—36
AtlasCloudfp80%$0.135$0.0135$0.54—40
Modalunknown0%$0.25$0.005$1.00—86
Mistralfp80%$0.30$0.007$1.20—52
Ambientfp40%$0.045$0.0025$0.25—22
StreamLakefp80%$0.147$0.00294$0.58851%44
GMICloudfp80%$0.18$0.0036$0.7240%89
DekaLLMunknown0%$0.10$0.003$0.60—41
Venicefp80%$0.30$0.0075$1.20—70
Googleunknown0%$0.30$0.03$1.20—42
Makorafp80%$0.26$0.0058$1.17—74
SiliconFlowfp80%$0.15$0.003$0.60—65
Tenstorrentfp80%$0.10$0.003$0.40—30
Inceptronfp40%$0.10$0.006$0.60—36
Ant Digitalfp80%$0.12$0.0024$0.4860%50
PrimaLabsunknown0%$0.24$0.0048$0.96—44
Near AIunknown0%$0.30$0.006$1.20—17

Left out, priced more than 4 times above or below the median host for this traffic: Mancer 2.

Every GLM-5.3 host (36)
HostShare of requestsInputCached inputOutputDiscountSpeed
Waferfp427%$0.0699$0.0689$1.40—56
Decartfp49%$0.56$0.10$1.81—66
InferenceNetfp48%$0.084$0.079$3.60—49
Togetherunknown7%$1.40$0.26$4.40—69
Tenstorrentfp86%$0.60$0.10$2.50—49
Z.AIfp85%$1.40$0.26$4.40—30
CoreWeavenvfp45%$0.75$0.19$3.00—63
DigitalOceanunknown4%$0.91$0.169$2.86—60
Coherenvfp43%$0.50$0.10$3.50—24
Sail Researchfp83%$0.75$0.12$3.00—41
Fireworksunknown2%$0.70$0.13$2.20—56
Friendliunknown2%$1.26$0.234$3.9610%66
Novitafp82%$0.63$0.117$1.9855%33
Cloudflareunknown2%$0.70$0.13$2.20—12
BaseTenfp42%$1.40$0.14$4.40—47
Nebiusfp41%$1.40$0.26$4.40—58
DeepInfrafp41%$0.563$0.125$2.5038%35
Inceptronfp41%$1.05$0.125$3.40—43
Mistralnvfp41%$1.54$0.154$4.84—21
Relacefp41%$0.0703$0.0703$4.38—44
Makorafp41%$1.50$0.26$4.40—96
Morphfp81%$1.04$0.212$4.88—46
ModelRunfp41%$2.10$0.40$6.60—76
Parasailfp41%$1.40$0.26$4.40—79
Modalunknown1%$1.40$0.26$4.40—57
Crusoefp40%$1.40$0.26$4.40—45
Amazon Bedrockunknown0%$1.68$0.312$5.28—52
GMICloudfp80%$0.70$0.13$2.2050%20
AkashMLfp40%$0.19$0.19$4.40—35
Featherlessunknown0%$0.42$0.078$1.3270%28
Veniceunknown0%$1.40$0.26$4.4020%52
Doublewordnvfp40%$0.25$0.13$1.00—35
AtlasCloudfp80%$1.40$0.26$4.40—19
SambaNovaunknown0%$1.40$0.07$4.4050%30
Alibabaunknown0%$1.19$0.238$3.74—34
Baidufp80%$1.40$0.26$4.40——
Every Kimi K3 host (20)
HostShare of requestsInputCached inputOutputDiscountSpeed
Wafermxfp418%$0.276$0.13$9.60—32
InferenceNetfp416%$0.245$0.214$15.00—48
Moonshot AIunknown14%$3.00$0.30$15.00—20
Nebiusfp49%$3.00$0.30$15.00—41
Togetherunknown8%$3.00$0.30$15.00—28
Fireworksunknown7%$3.00$0.30$15.00—26
Relacefp46%$0.18$0.18$11.01—42
DeepInframxfp45%$2.85$0.285$14.25—46
Riverunknown4%$3.00$0.30$15.00—74
Sail Researchfp43%$0.345$0.172$16.58—39
Parasailfp43%$3.00$0.30$15.00—27
KTunknown2%$0.45$0.30$15.00—26
Makorafp42%$1.00$0.50$18.00—36
Io Netfp41%$1.90$0.19$14.00—45
Amazon Bedrockunknown1%$3.30$0.33$16.50—42
BaseTenfp81%$3.00$0.30$15.00—32
Modalmxfp41%$3.00$0.30$15.00—46
Alibabaunknown1%$3.00$0.30$15.00—26
DigitalOceanunknown0%$2.55$0.255$12.95—30
Chutesmxfp40%$1.85$0.185$9.50—18
Every GLM-5.3-Flash host (35)
HostShare of requestsInputCached inputOutputDiscountSpeed
CoreWeavenvfp424%$0.12$0.02$0.38—53
Coherenvfp410%$0.105$0.02$0.34—52
Wafernvfp49%$0.042$0.041$0.184—60
Togetherunknown7%$0.15$0.03$0.50—92
DigitalOceanunknown7%$0.113$0.03$0.375—61
Fireworksunknown5%$0.15$0.03$0.50—56
Relacefp44%$0.0399$0.03$0.0784—32
Particle AInvfp44%$0.13$0.04$0.50—70
Z.AIfp84%$0.15$0.03$0.50—24
InferenceNetfp43%$0.108$0.035$0.35—57
OpenInferencefp43%$0.05$0.025$0.45—82
Novitafp83%$0.084$0.0168$0.2844%10
DekaLLMunknown2%$0.10$0.03$0.50—16
BaseTenfp82%$0.15$0.03$0.50—57
Riverunknown2%$0.12$0.024$0.40—54
Friendliunknown2%$0.15$0.03$0.50—60
Modalfp42%$0.15$0.03$0.50—67
Near AIfp81%$0.08$0.012$0.50—9.0
DeepInfrafp41%$0.101$0.0201$0.33533%46
Makorafp41%$0.14$0.03$0.24—59
StreamLakefp81%$0.0885$0.0177$0.29541%16
Rekaunknown1%$0.12$0.03$0.30—43
Mistralunknown1%$0.15$0.03$0.50—60
GMICloudfp81%$0.075$0.015$0.2550%21
Inceptronfp81%$0.16$0.04$0.50—47
Crusoefp40%$0.15$0.03$0.50—62
Parasailfp80%$0.15$0.03$0.50—10
Morphfp80%$0.103$0.052$0.298—24
Tenstorrentfp80%$0.075$0.015$0.25—23
KTunknown0%$0.15$0.03$0.50—37
Sail Researchfp40%$0.175$0.0225$0.625—44
Veniceunknown0%$0.15$0.03$0.50—51
Darkbloomfp40%$0.065$0.012$0.18—2.0
Nebiusfp40%$0.15$0.03$0.50—53
Baidufp80%$0.15$0.03$0.50—24

Yields

From SemiAnalysis InferenceX’s agentic benchmark: the best official configuration per GPU at the model’s target speed per user (p90), from its latest run. Total includes cached input; the crack prices cached and uncached input separately.

ModelGPUOutput tok/sTotal tok/sCachedConfigurationRun
V4.1-Flashat 65 tok/s per userH100130†20K93%vllm, fp4† measured only at 68 tok/s per user and fasterNov 20
V4.1-Flashat 65 tok/s per userH20032852K97%vllm, fp4Nov 20
V4.1-Flashat 65 tok/s per userB2002,800†388K95%vllm, fp4† measured only at 107 tok/s per user and fasterNov 20
V4.1-Flashat 65 tok/s per userB30013,908†828K98%dynamo-vllm, fp4† measured only at 84 tok/s per user and fasterNov 20
V4.1-Flashat 65 tok/s per userGB20013,354†817K96%dynamo-vllm, fp4† measured only at 126 tok/s per user and fasterNov 20
V4.1-Flashat 65 tok/s per userGB30015,785†1.17M91%dynamo-vllm, fp4† measured only at 142 tok/s per user and fasterNov 20
V4.1-Flashat 65 tok/s per userMI300X14521K96%vllm, fp4Nov 20
GLM-5.3at 45 tok/s per userB200298†20K94%dynamo-sglang, fp4† measured only at 86 tok/s per user and fasterNov 18
GLM-5.3at 45 tok/s per userB30024037K95%dynamo-sglang, fp4Nov 18
GLM-5.3at 45 tok/s per userGB200768†42K93%dynamo-sglang, fp4† measured only at 92 tok/s per user and fasterNov 18
GLM-5.3at 45 tok/s per userGB3001,145†64K88%dynamo-sglang, fp4† measured only at 89 tok/s per user and fasterNov 18
Kimi K3at 25 tok/s per userB2007310K95%dynamo-vllm, fp4Nov 4
Kimi K3at 25 tok/s per userB300638,33391%vllm, fp4Nov 4
Kimi K3at 25 tok/s per userGB2007611K94%dynamo-vllm, fp4Nov 4
Kimi K3at 25 tok/s per userGB30010†1,56497%dynamo-vllm, fp4† measured only at 433 tok/s per user and fasterNov 4

GLM-5.3: InferenceX benchmarks GLM-5.2 and GLM-5.3 as one model: they share an architecture.

Method

The crack. For each GPU and model: token value per GPU-hour = 3,600 × (output tokens a second × the output price + input tokens a second × (uncached share × the input price + cached share × the cached-input price)) ÷ 1,000,000, with prices per million tokens; crack = token value − rent. The break-even load is rent ÷ token value: the share of the benchmark’s output the GPU must sustain to cover its rent. The owner’s crack uses the cost of owning instead of rent. “Pays at 50% load” is token value × 0.5.

Yields. For each model and GPU, the highest official throughput per GPU among the configurations InferenceX runs (engine, precision, parallelism) at the model’s target speed per user, from InferenceX’s runs in the 14 days to the day, so a run that covers only some configurations doesn’t knock a GPU’s yield down. A configuration measured only at speeds below the target is left out, since its throughput there would overstate what the GPU does at the target. One measured only at speeds above the target counts at its slowest measured point, which understates it; those yields are flagged. InferenceX’s agentic benchmark (AgentX): replayed agentic coding sessions, multi-turn and long-context, with most of each request’s input already in the cache from earlier turns. Each result gives input and output tokens a second per GPU and the share of input served from cache, and those are what the crack prices. Each model’s target is the median p90 speed of its standard OpenRouter hosts, weighted by their requests, rounded to the nearest 5 tok/s. It is set at the first collection of each calendar month and holds for that month. p90: the speed 90% of requests reach or beat, the measure InferenceX uses for agentic work. Yields change only when InferenceX publishes a new run, and on that day the crack picks it up; the daily series shows how much of each move came from yields.

Token prices. From OpenRouter’s list of every host serving each model. Prices are what each host charges, net of any discount it advertises. The realized price of each kind of token is the average over the model’s standard endpoints weighted by the requests each served in the 24 hours to the fixing (OpenRouter reports requests in the last 30 minutes at every collection). Without request data it falls back to OpenRouter’s default routing, which weights hosts by the inverse square of their price. The median host’s posted prices are kept beside it. Left out: free endpoints, faster or cheaper service tiers (priority, flex), hosts OpenRouter marks as degraded or down, and endpoints whose price for the model’s traffic mix is more than four times above or below the median host’s. A model needs at least 5 priced hosts for a print.

Rents. The inthesingularity.ai GPU Compute Price Index, global, all providers: spot and on-demand indices, and commitments of a month, a year and three years at on-demand times each provider’s own term discount, the median across providers (pooled across GPUs where fewer than three providers quote a GPU’s term). GB300 isn’t indexed yet, so its rent is provisional: each provider’s on-demand price, or its shortest commitment where it posts none, from the index database, and the median across providers. Cost of owning: SemiAnalysis’s total cost of ownership per GPU-hour, as published in InferenceX.

Models and rolls. Each series follows a class of model: efficient (the most-used open-weight model on OpenRouter: DeepSeek V4.1-Flash), large (a large open model, several hundred billion parameters: GLM-5.3) and frontier-size (the most-used open model of a trillion parameters or more: Kimi K3). Each series follows a class rather than a model, like a continuous futures contract. A class moves to a successor once the successor has carried more paid tokens on OpenRouter than the current model for seven straight days and InferenceX has agentic results for it on at least four of the GPUs here. The move takes effect at the next fixing, is logged below and is marked on the charts. Also collected, without a crack: GLM-5.3-Flash. The second most-used open model on OpenRouter. InferenceX doesn’t benchmark it, so its prices are collected but it has no crack.

Fixing. One a day at 16:00 UTC, with the GPU index’s fixing, from the token prices collected at or before then (within 36 hours), InferenceX’s latest run on or before the day, and the index’s rents. A final fixing stands as published. Between fixings, the figures above show the same calculation on the latest prices.

What it leaves out. Power and networking where they’re billed separately, staff and idle time. InferenceX measures tuned configurations at full load, and real fleets run below it, which is why the break-even load matters.

Updates. Prices are collected through the day with the GPU index, and InferenceX is checked at every collection. Download every daily fixing and the latest print: gpu-crack-daily.csv. The raw collections (every host’s prices, traffic and speed, every InferenceX result) are in the site’s database.

Changes. Oct 27, 2026: First version: yields from NVIDIA’s Dynamo recipes and one vLLM run, a standard 64K-input trace, median host prices, on-demand rent. Oct 28, 2026: Before the first fixing, the method changed. Yields now come from SemiAnalysis InferenceX by rule (one benchmark for every GPU, at the speed hosts deliver), so H100 and B300 get cracks. Token prices are weighted by the requests each host serves, with the median host’s posted prices kept beside them. GLM-5.3 replaces GLM-5.3-Flash, which InferenceX doesn’t benchmark. MI300X added. New measures: cracks against commitment rents, the owner’s crack, the highest rent a host can pay at half load, and each series follows a model class that rolls to successors.

Yields: SemiAnalysis InferenceX, which describes its data as open source and freely available; credited wherever used. Cost of ownership: SemiAnalysis, as published in InferenceX. Token prices, traffic and speeds: OpenRouter’s listings of each host. Rents: the inthesingularity.ai GPU Compute Price Index. Cracks are inthesingularity.ai calculations. Not investment advice. See also the Who makes money serving open models? analysis.