Hire LLM Engineers in the Philippines

What LLM Engineers Earn in the Philippines Working for International Clients

The Philippine floor sits $220 below India's, and China's range opens $4,890 above the point where the Philippine card turns open-ended. We publish no LLM engineer card, so the table uses our AI engineer rate cards, the closest match. The figures are what engineers in each market earn working directly for foreign companies, before any employer or platform costs.

Monthly ranges from the Remote (Working for International Clients) figures on our rate cards for the Philippines, India, Vietnam and China, converted at ExchangeRate-API mid-market rates for 14 September 2026.

The Philippine card quotes ₱150,000 to ₱300,000+ a month, converted at ₱62.69 to the dollar.

For an engineer who trains and fine-tunes models in PyTorch rather than calling hosted ones, our Philippines machine learning engineer card runs $2,500 to $6,000, quoted in US dollars.

GPU time and token spend come on top of pay for this role. Put a compute budget in the brief next to the salary line.

Through Second Talent you pay one monthly fee that bundles salary, payroll taxes, statutory contributions and our service fee, quoted per role and market. Our pricing page explains the monthly subscription, and for short projects our LLM developer cost-to-hire page puts a mid-level US freelance or contract LLM developer at $119 to $185 an hour.

US Pay Benchmark for LLM Engineers

An LLM engineer who builds products on models maps to Software Developers (SOC 15-1252). One who runs training, fine-tuning experiments and evaluation research sits closer to Computer and Information Research Scientists (15-1221). The table shows both.

Monthly figures are the annual wages on the BLS OEWS profiles for 15-1252 and 15-1221 divided by 12, rounded to the nearest $10.

The medians sit $360 a month apart, and the gap widens to $1,330 at the 90th percentile. Use the research column for a hire who will own post-training experiments.

Salary is only part of the US cost. In the BLS Employer Costs for Employee Compensation release for June 2026, wages and salaries made up 68.5% of employer compensation costs for full-time private industry workers, and benefits the other 31.5%.

Time Zone Overlap: Manila Against ET, CT and PT

Manila's midnight is 9:00 am in San Francisco for most of 2026. The city stays on UTC+8 with no daylight saving time, and US daylight time runs from 8 March to 1 November 2026, per NIST.

The arithmetic: 9:00 am under PDT (UTC-7) is 16:00 UTC, and 16:00 UTC is 12:00 midnight in Manila. Under PST (UTC-8) it becomes 1:00 am. For Chicago, 9:00 am CDT (UTC-5) is 14:00 UTC, or 10:00 pm in Manila.

Each shift is nine clock hours, including a meal break, counted against a 9 am to 5 pm US day.

Employees earn a night shift differential of at least 10% for each hour between 10 pm and 6 am, under Article 86 of the Labor Code. The 10 pm to 7 am row pays it on eight hours.

LLM work suits the 5 pm to 2 am row. An engineer reviews eval results with an East Coast team in its morning, then leaves a fine-tuning job or a batch evaluation running through the US afternoon and the Manila night.

The 1 pm to 10 pm row gives almost no live time and fits only a team that works in writing.

We agree the shift before the offer, which is how our placements get 4-6 hours of daily overlap with US hours.

LLM Engineer Skills to Screen For

In Stack Overflow's 2025 survey, 81.4% of respondents who named models for development work used OpenAI's GPT models, 17.8% used Meta Llama and 5.2% used Alibaba's Qwen. Find out which families the candidate has shipped on, then test the five areas below.

Evaluation in Tagalog and Cebuano

FilBench, published on Hugging Face in August 2025, splits 12 tasks into four categories: cultural knowledge, classical NLP, reading comprehension and generation. Its authors found that Southeast Asian models often scored highest for their size. The best of them still trailed GPT-4o. Continued fine-tuning on regional instruction data added 2-3%.

Most models struggled with generation. The failures included ignoring translation instructions, writing overly long answers and switching to another language instead of Tagalog or Cebuano. Ask the candidate to build a small eval set for a Taglish support bot and explain how they would detect a reply in the wrong language.

PyTorch after Transformers v5

The v5 post counts more than 3 million Transformers installs a day and over 400 model architectures. PyTorch is the only backend in v5, so you have to port a TensorFlow fine-tuning script before it runs on the new release.

Hand the candidate a short TensorFlow fine-tuning loop and ask how they would port it to PyTorch, what they would test, and which checkpoint formats carry across.

Adapters before full fine-tuning

LoRA cut trainable parameters 10,000 times and GPU memory three times against full fine-tuning of GPT-3 175B, with no added inference latency. Give the candidate a labelled dataset and a prompt baseline. Score whether they test a better prompt and retrieval before training anything, and report cost per point of improvement.

Serving open-weight models

Meta's Llama 4 Maverick repository holds about 400 billion parameters in BF16, over 800 GB of files. The vLLM paper showed that PagedAttention raised throughput two to four times at the same latency by cutting wasted key-value cache memory.

Licence terms travel with the weights. The Llama 4 licence requires anyone who distributes or makes available a product containing Llama materials to display "Built with Llama" on a related website, interface or documentation. A business that had more than 700 million monthly active users on the release date must ask Meta for a licence.

Ask the candidate to size GPUs for your peak traffic and name the licence clause that applies to your product.

Long context and prompt injection

Chroma's context rot study tested 18 models, including GPT-4.1, Claude 4, Gemini 2.5 and Qwen3, and found that performance "grows increasingly unreliable as input length grows."

OWASP's LLM01 entry adds that content from websites or files can carry indirect prompt injections that change the model's behaviour.

Ask how the candidate decides what goes into a context window, and how a retrieval pipeline stops a poisoned document from issuing instructions.

Contractor or Employer of Record in the Philippines

Software is not one of the nine categories of commissioned work that can be "work made for hire" under 17 U.S.C. § 101, and a transfer must be in writing and signed under § 204(a).

Under Section 178.4 of the Philippine Intellectual Property Code, the party that commissions a work owns it, but the copyright stays with the creator unless a written stipulation says otherwise. Section 178.3 gives an employer the copyright in work from an employee's regularly assigned duties. List LoRA adapters, merged weights, eval sets, prompts, training scripts and data pipelines in the assignment.

IRS Publication 515 says the place where services are performed determines the source of the income, so an LLM engineer working from Manila or Cebu earns foreign-source income. A foreign individual gives the payer Form W-8BEN to certify foreign status. On the Philippine side, BIR Revenue Regulations 8-2018 give self-employed individuals below the VAT threshold the option of an 8% tax on gross receipts above ₱250,000.

Philippine courts apply the four-fold test: selection and engagement, payment of wages, the power of dismissal and control over how the work is done. In Atok Big Wedge v. Gison, the Supreme Court called control the most important. A contractor whose experiments your ML lead assigns and reviews each day starts to look like an employee.

An employee adds four statutory lines to salary:

• 13th month pay: one twelfth of basic salary, paid by 24 December under PD 851 and Memorandum Order 28.
• SSS: 15% of the monthly salary credit, shared by employer and employee, with the credit capped at ₱35,000.
• PhilHealth: 5% of monthly income, capped at ₱100,000.
• Pag-IBIG: 2% from each side on a salary capped at ₱10,000, per the February 2024 increase.

Our Employer of Record service carries those lines, and our Philippines EOR page and Philippines payroll guide show the employer cost. This is a summary, not legal advice.

English Level and Working Norms in the Philippines

The Philippines' national score of 569 on the EF English Proficiency Index 2025 ranks 28th of 123 countries and regions, against a global average of 488. Its IT job-function score is 581, and writing, at 603, leads speaking at 539, per EF's Philippines page.

An LLM engineer writes much of what they deliver: system prompts, eval rubrics, model cards and failure reports. Ask finalists to write the model card for their assessment fine-tune, covering training data, eval scores and known failures.

Proclamation 1006 makes Christmas Day and Rizal Day on 30 December regular holidays in 2026, with 24 and 31 December special non-working days. A training run launched before that week needs someone watching it, so agree cover before the break.

Thanksgiving on 26 November 2026 is a working day in the Philippines, so a Manila engineer can finish an evaluation cycle while your US team is off.

LLM Engineer Hiring Process in the Philippines

Second Talent's vetting covers stages 2 to 5 before you see a profile.

Stage 1: Define what the hire owns You state whether the engineer calls hosted models, fine-tunes open-weight models or serves them on your GPUs. Name the languages your users write in, the licences you accept and the compute budget, and add the Manila shift from the table above.

Stage 2: Application review We look for a model in production with written evaluation: a fine-tune with before-and-after scores, a retrieval system with measured recall, or an inference stack under real traffic. A notebook demo without numbers does not pass.

Stage 3: Skills assessment The candidate improves a baseline on a small English and Tagalog dataset with a better prompt, retrieval or a LoRA adapter, written in PyTorch. They report scores by language, cost per run and the licence of each model used.

Stage 4: Live technical interview with a senior engineer A senior engineer reviews the report with the candidate in English, then moves to serving and safety: GPU sizing for your traffic, a long prompt that degrades answers, and a document that tries to inject instructions.

Stage 5: Background and reference checks We ask former managers how the candidate's models held up after release, how they handled training data and how they reported failures. You then choose the contractor or EOR route above.

Hire LLM Engineers from the Philippines with Second Talent

We shortlist 6-8 candidates within 24 hours from 100,000+ pre-vetted engineers, accepting only the top 1% of applicants. LLM engineers from the Philippines come with $0 upfront, no lock-in and 4-6 hours of daily overlap with US hours. We handle contracts, payroll and equipment, with compliant EOR contracts in the Philippines.

Our pricing page sets out the monthly subscription, and for product AI work we also staff AI developers in the Philippines.