Crusoe Launches Serverless Fine-Tuning and Self-Serve Inference Deployments, Accelerating Open-Model Development From Experiment to Production

SAN FRANCISCO, July 07, 2026 (GLOBE NEWSWIRE) — C usoe, the wo ld’s fi st ve tically i teg ated AI i f ast uctu e compa y, today a ou ced Se ve less Fi e-Tu i g a d Self-Se ve Deployme ts i C usoe I tellige ce Fou d y, the ma aged AI platfo m fo C usoe Cloud. Togethe , these capabilities give data scie tists a d ML e gi ee s a complete path f om p op ieta y data to p oductio – eady models—o pu pose-built AI i f ast uctu e, without the ove head of ma agi g it.

Fi e-tu i g is ow a sta da d pa t of buildi g with ope -sou ce AI models — a d as ope -weight models catch up to p op ieta y models, mo e teams a e choosi g to customize  with thei p op ieta y data a d etai ow e ship of the fi e-tu ed weights. Getti g sta ted is st aightfo wa d, but doi g it epeatedly adds up. Idle cluste s, ha dwa e failu es, a d scatte ed tools slow teams dow , a d the people who should be imp ovi g the model e d up t oubleshooti g i f ast uctu e i stead.

C usoe Se ve less Fi e-Tu i g: F om Data to Deployed Model i Mi utesC usoe Se ve less Fi e-Tu i g elimi ates i f ast uctu e ove head so e gi ee i g teams ca focus o model quality. Teams ca lau ch a fi e-tu i g job i a few clicks — select a base model f om a cu ated lib a y of top-pe fo mi g ope -weight models, upload a custom data set, co figu e setti gs with p e-co figu ed best p actices, a d submit; o dedicated ese vatio equi ed. Jobs u o C usoe’s dist ibuted AI-optimized i f ast uctu e with automated ecove y a d esta t if ha dwa e blips a e detected. As soo as the model stops imp ovi g, teams stop payi g. Whe tu i g is complete, model weights a e etu ed i po table .safete so s fo mat. They ca go live i o e click usi g C usoe’s ew Self-Se ve Deployme ts fo i fe e ce, o the weights ca be dow loaded to deploy a ywhe e.

“Ou ea ly expe ie ce with C usoe’s Se ve less Fi e Tu i g p oduct was seamless, a d it wo ked like a cha m. We look fo wa d to leve agi g it to optimize the late cy a d cost of ou AI age ts as we scale ou i f ast uctu e.” — D . Will Lee ey, D . Hiskias Di geto, AI Resea che s, StackO e

“Ope models have defi itely c ossed the quality th eshold, while offe i g u ique optimizatio oppo tu ities with you data, a d givi g you full co t ol of thei lifecycle, ” said E wa Me a d, Se io Vice P eside t of P oduct, C usoe Cloud. “With C usoe Se ve less Fi e-Tu i g a d Self-Se ve Deployme ts you jou ey just got easie ; fast ite atio , p edictable cost, a d the gua a tee that you data a d weights stay you s. You should ‘t have to choose betwee a ma aged expe ie ce a d ow e ship of you model.”

C usoe Self-Se ve Deployme ts: F om Fi e-Tu i g to Flexible I fe e ce i O e ClickC usoe Cloud is expa di g the co sumptio optio s available fo C usoe Ma aged I fe e ce a d addi g Self-Se ve Deployme ts. While Se ve less I fe e ce APIs a e ideal fo ea ly-stage expe ime tatio , Self-Se ve Deployme ts is desig ed fo teams with p oductio – eady wo kloads. Billed by GPU pe hou fo p edictable costs, use s ca select a base model i I tellige ce Fou d y, choose a i fe e ce p ofile optimized fo th oughput o espo sive ess, a d deploy to p oductio o NVIDIA H100 o H200 GPUs—without eve touchi g the u de lyi g i f ast uctu e. Fo teams u i g co ti uous post-t ai i g loops with Se ve less Fi e-Tu i g i I tellige ce Fou d y, the ability to tu e models a d deploy to a p oductio i fe e ce e dpoi t i o e click elimi ates the f ictio of movi g model a tifacts betwee p ovide s.

Self-Se ve Deployme ts expa ds the i fe e ce optio s available i C usoe I tellige ce Fou d y. Custome s ca ow choose Se ve less I fe e ce APIs fo quick expe ime tatio , Self-Se ve Deployme ts fo p oductio – eady wo kloads optimized fo th oughput o espo sive ess, o Tailo ed Deployme ts fo dedicated, custom i fe e ce o a y fi e-tu ed o p op ieta y model with SLA-backed pe fo ma ce.

Teams buildi g the ext ge e atio of AI p oducts — like Yuto i, Nous Resea ch, Wo de ful, Salie t, Composite, a d Magica e — u o C usoe because fast, eliable i fe e ce optimized to thei stack is how they delive best-i -class pe fo ma ce to the people who depe d o them.

Key Capabilities at Ge e al Availability

Se ve less Fi e-Tu i g:

• Develope -f ie dly UI, SDK, a d API
• Cu ated lib a y of top pe fo mi g base model families, i cludi g Qwe , DeepSeek, Gemma, gpt-oss, a d mo e
• LoRA (low- a k adaptatio ) fi e-tu i g fo lightweight customizatio that delive s fast, cost-efficie t ite atio
• Automatic job ecove y a d esta t, plus checkpoi ts that save at eve y step – ea ly stoppi g e ds billi g the mome t the model stops imp ovi g
• Full job li eage: eve y tu ed a tifact t aces back to the exact data a d co figu atio that p oduced it
• Native expo t of aw weights i .safete so s fo mat

Self-Se ve Deployme ts:

• P edictable pe fo ma ce fo p oductio -g ade wo kloads
• I fe e ce p ofiles optimized fo th oughput o espo sive ess based o you u ique eeds
• Ope AI-compatible API fo ze o-f ictio i teg atio with existi g applicatio s
• O e-click deployme t f om Se ve less Fi e-Tu i g wo kflow

P ici g a d Availability

Se ve less Fi e-Tu i g a d Self-Se ve Deployme ts will be ge e ally available ext week i C usoe I tellige ce Fou d y. P ici g fo Se ve less Fi e-Tu i g is toke -based, p iced pe o e millio toke s p ocessed. Self-Se ve Deployme ts a e billed by GPU hou . Fo both Se ve less Fi e-Tu i g a d Self-Se ve Deployme ts, teams ca optio ally co t act fo mo thly o volume ates.

About C usoe

As the AI facto y compa y, C usoe is o a missio to accele ate the abu da ce of e e gy a d i tellige ce. The compa y p ovides a eliable, scalable, cost-effective, e e gy-fi st solutio fo AI i f ast uctu e. By ha essi g la ge-scale e e gy sou ces, buildi g AI-optimized data ce te s, a d delive i g a powe ful AI cloud platfo m, C usoe empowe s its custome s a d pa t e s to build the futu e faste . Fo mo e i fo matio , visit c usoe.ai.

Media co tact

media@c usoe.ai