PERFORMANCEGLM-5.2 → Scaleway: TTFT ↑ 20%28mPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↑ 35%28mPERFORMANCEMistral Medium 3.5 → Scaleway: TTFT ↓ 21%28mPERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↓ 20%28mPERFORMANCEGPT-OSS 120B → Scaleway: throughput ↑ 31%28mPERFORMANCEGPT-OSS 120B → Scaleway: TTFT ↓ 24%28mPERFORMANCEGPT-OSS 120B → Together AI: throughput ↑ 20%28mPERFORMANCEKimi K3 via OpenRouter: TTFT ↓ 49%28mPERFORMANCEGLM-5.2 via OpenRouter: reliability recovered 96.8% → 100.0%28mPERFORMANCEKimi K3 via Cortecs: throughput ↑ 184%28mPERFORMANCEKimi K3 via Cortecs: TTFT ↓ 42%28mPERFORMANCEGLM-5.2 via OpenRouter: throughput ↓ 49%28mPERFORMANCEGLM-5.2 via OpenRouter: TTFT ↑ 121%28mPERFORMANCEMiniMax M3 via OpenRouter: reliability recovered 96.8% → 100.0%28mPERFORMANCEDeepSeek V4 Pro via Cortecs: throughput ↑ 27%28mPERFORMANCEGPT-OSS 120B via OpenRouter: throughput ↓ 22%28mPERFORMANCEGLM-4.7 via OpenRouter: TTFT ↓ 78%28mPERFORMANCEGPT-OSS 20B → Groq: TTFT ↑ 24%28mPERFORMANCEKimi K3 → Together AI: TTFT ↓ 34%28mPERFORMANCEDeepSeek V4 Flash via OpenRouter: throughput ↓ 17%28mPERFORMANCELlama 3.3 70B via Cortecs: TTFT ↓ 51%28mPERFORMANCEQwen3 235B via OpenRouter: throughput ↑ 37%28mPERFORMANCEQwen3 235B via OpenRouter: TTFT ↓ 29%28mPERFORMANCEDeepSeek V4 Flash → Together AI: throughput ↑ 19%28mPERFORMANCEGLM-5.3 via OpenRouter: throughput ↓ 17%28mPERFORMANCEGPT-OSS 20B via Cortecs: throughput ↑ 28%28mPERFORMANCELlama 3.3 70B → Together AI: TTFT ↓ 33%28mPRICE~moonshotai/kimi-latest cached_input_per_mtok decreased from 0.80 to 0.2929mPRICE~moonshotai/kimi-latest output_per_mtok decreased from 13.00 to 11.3629mPRICE~moonshotai/kimi-latest input_per_mtok decreased from 0.99 to 0.7529mPRICE~deepseek/deepseek-v4-flash-latest cached_input_per_mtok decreased from 0.0077 to 0.001129mPRICE~deepseek/deepseek-v4-flash-latest output_per_mtok decreased from 1.28 to 1.0429mPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok decreased from 0.0077 to 0.003829mPRICEz-ai/glm-5.1 cached_input_per_mtok increased from 0.18 to 0.2629mPRICEz-ai/glm-5.1 output_per_mtok increased from 3.03 to 4.4029mPRICEz-ai/glm-5.1 input_per_mtok increased from 0.96 to 1.4029mPRICEtencent/hy4-preview cached_input_per_mtok increased from 0.038 to 0.04229mPRICEtencent/hy4-preview input_per_mtok increased from 0.75 to 0.8329mPRICEtencent/hy3 cached_input_per_mtok increased from 0.021 to 0.03329mPRICEtencent/hy3 output_per_mtok increased from 0.33 to 0.5329m

InferenceBench Partners

Put your brand beside the independent benchmark engineers use to choose AI inference infrastructure.

5 founding positions · partnerships scoped to each companyBecome a Partner

Partner Program Terms & Independence Policy →

Founding Partner Program

Five founding partnerships for companies building the next generation of AI infrastructure. Founding partners help shape the program, receive early access to new benchmark suites and direct product collaboration — and hold a founding mark no later partner can carry.

FOUNDING PARTNER
RESERVED
RESERVED
RESERVED
RESERVED

Partners support InferenceBench financially. They have no influence over benchmark methodology, scores, rankings or editorial conclusions.

Where your brand appears

Every panel below is the live product in miniature. Your mark renders exactly where the placeholder sits — beside independent data, on every visit, never inside it.

inferencebench.dev — live ranking

The benchmark workspace

The partner rail beside the live ranking — present on the first screen of every benchmark session, clearly separated from the data.

/providers/* · /models/*

Provider & model dossiers

Presence on selected provider and model dossiers — beside the measurements engineers compare when they choose infrastructure.

/research

Research reports

Clearly disclosed support on published research — read by the teams your developer marketing wants to reach, under our methodology alone.

What a Benchmark Partner receives

Product visibility
  • Logo and link in the Benchmark Partner rail
  • Presence on selected provider and model detail surfaces
  • Inclusion on this page
  • Acknowledgement in relevant releases
Partner intelligence
  • Periodic competitive position report
  • Per-path measured metrics with confidence and eligible-result counts
  • Market medians across every measured path, same vantage
  • Published events and exclusion accounting for the window
Research
  • Partner briefings and relevant research distribution
  • Participation in selected technical studies
  • Clearly disclosed supported research where applicable
Custom benchmarking
  • Agreed independent custom benchmark capacity
  • Private studies where appropriate
  • Public studies always under standard InferenceBench methodology
Data
  • Extended historical dataset — available on selected partnerships
  • API access and benchmark exports — available on selected partnerships

Partner Intelligence

Partners receive a continuously measured view of their own position — the same engine that powers the public benchmark, focused on you. A sample positioning report:

Your InferenceBench position

Sample · DEV DATA
  • Overall#5 of 8
  • ReliabilityP38
  • Tool callingP13
  • Structured outputP50
  • TTFTP50
  • Price / performanceP63

Intelligence helps partners improve — it never reveals benchmark corpora, unpublished weights or anything that would let a result be gamed.

Independent custom benchmarks

Need to evaluate a new endpoint, model deployment, region or infrastructure configuration? InferenceBench designs and executes independent benchmark studies — private for engineering and competitive intelligence, or public under the standard methodology.

Commissioning the benchmark buys the study, not the result.

Available to partners and selected infrastructure companies.

Independence is not for sale.

Partners support InferenceBench financially. They cannot purchase ranking position, influence methodology, change scores, suppress results or control editorial conclusions.

What partnership does not buy—
Ranking positionNever
Score adjustmentsNever
Methodology influenceNever
Preferential benchmark treatmentNever
Removal of unfavorable resultsNever
Editorial controlNever
Guaranteed positive coverageNever

Who should partner

The program is not limited to ranked providers — partners fund the benchmark; they do not need to appear in it.

  • AI inference providers
  • GPU cloud platforms
  • AI infrastructure companies
  • Model serving platforms
  • AI gateways
  • Observability & evaluation companies
  • Developer infrastructure
  • Sovereign cloud providers
  • Data center & compute infrastructure
  • Hardware ecosystem companies

Start a partnership conversation

Tell us who you are and what you want to achieve — we'll come back with a scoped proposal.

What are you hoping to achieve? *

SnowStorm Solutions S.L. will use the information you provide to respond to your partnership enquiry. Privacy Policy → Partner Terms →

Prefer email? partners@inferencebench.dev

Partners support InferenceBench financially. They have no influence over benchmark methodology, scores, rankings or editorial conclusions.