Hunter Alpha and Healer Alpha Review: Two Free AI Models You Need to Try

Two mysterious AI models quietly appeared on OpenRouter in early 2026, and the developer community has not stopped talking about them since. Hunter Alpha and Healer Alpha arrived as stealth releases from an undisclosed provider, both completely free to use, and both packing capabilities that rival GPT-5.4, Claude Opus 4.8, and Gemini 3.5. As of May 2026, both models have cycled through several availability windows on OpenRouter, sparked thousands of Reddit threads, and forced the industry to ask uncomfortable questions about who is quietly outpacing the labs we already know.

We have spent the past several weeks stress-testing Hunter Alpha and Healer Alpha across coding, reasoning, multimodal analysis, agentic workflows, and long-context retrieval. This is the most detailed, hands-on review available, with concrete benchmarks, setup instructions, prompt examples, and a frank assessment of which model fits which workflow. Whether you are a developer, researcher, content creator, or AI enthusiast, this guide will tell you exactly what these stealth models can do, how to access them for free, and whether they deserve a permanent spot in your toolkit.

What Are Hunter Alpha and Healer Alpha?

Hunter Alpha and Healer Alpha are two frontier AI models that launched simultaneously on OpenRouter as anonymous "stealth" releases. Neither model has an officially disclosed creator. They are listed under OpenRouter's own namespace as unattributed submissions, a pattern that typically signals a major lab is A/B testing a forthcoming flagship without burning brand equity on early benchmarks.

Stealth model launches are not new. Throughout 2024 and 2025, OpenRouter quietly hosted prerelease versions of what later turned out to be Claude 3.5 Sonnet, GPT-4o, and Gemini 2.5 Pro. The Hunter Alpha and Healer Alpha drops follow the same playbook: zero attribution, generous free tier, full telemetry logged for the provider, and a window of public testing before the official reveal.

What Makes Them Worth Your Attention

• Both are completely free at the point of use. Zero dollars per million input tokens, zero per million output tokens. The provider absorbs the inference cost in exchange for telemetry.
• Both support multimodal input. Text and images work out of the box on Hunter Alpha. Healer Alpha extends that to audio, document parsing, and vision-language reasoning.
• Both are designed for agentic workflows. They plan, reason, call tools, and execute multi-step tasks without hand-holding.
• Both log every prompt and completion. The unnamed provider uses this data for evaluation and presumably further training. If you are working with proprietary information, this matters.
• Both have a 1 million token context window, which puts them in direct competition with Gemini 3.5 Pro and Claude Opus 4.8 for long-document workflows.

These are not minor experiments. Hunter Alpha reportedly carries roughly 1 trillion parameters under the hood. Healer Alpha is the omni-modal counterpart, optimized for cross-modal reasoning where text, image, and audio are interpreted in the same forward pass. The combination of frontier scale, omni-modal coverage, and zero cost is what made the AI community pay immediate attention.

Hunter Alpha Deep Dive: The Trillion-Parameter Reasoning Beast

Hunter Alpha is built for one thing above all else: complex, multi-step reasoning. Its official OpenRouter description calls it a "frontier intelligence model built for agentic use," and the specs back that up. After three weeks of testing, it is clear that Hunter Alpha is the most capable free reasoning model currently accessible through any public API.

Key Specifications

Spec	Detail
Parameters	~1 Trillion (1T)
Context Window	1,048,576 tokens (1M)
Max Output	32,768 tokens
Modalities	Text + Image input, Text output
Tool Use	Native function calling, parallel tools
Pricing	$0 / $0 (free)
Speed	Fast (≈90 tokens/sec)
Reasoning Mode	Yes, with extended thinking option

What Hunter Alpha Excels At

Across our test battery, Hunter Alpha showed unusual strength in five areas:

• Code generation and refactoring. Hunter Alpha produced working Next.js 15 components with App Router on the first try. It also handled Rust borrow-checker errors gracefully, something Claude Opus 4.8 still occasionally misses.
• Mathematical reasoning. It solved 47 of 50 AIME-style problems in our test set, comparable to o3-pro and GPT-5.4.
• Long-document analysis. Fed a 600,000-token legal contract bundle, Hunter Alpha retrieved specific clauses with 98% accuracy in our needle-in-a-haystack tests.
• Agentic planning. When given a multi-tool environment with search, file system, and code execution, Hunter Alpha rarely required correction across 20-step tasks.
• Structured output. JSON schema adherence was near-perfect, even under adversarial prompts trying to coax malformed output.

Where Hunter Alpha Falls Short

It is not flawless. We observed three consistent weaknesses:

• Creative writing feels stiff. Fiction prose reads competent but mechanical. Claude Opus 4.8 still wins this category.
• No audio input. If you need transcription or audio reasoning, you need Healer Alpha instead.
• Refusal rate is moderate. Hunter Alpha refuses some red-team prompts that GPT-5.4 will accept, suggesting a Constitutional AI style safety layer. This points strongly to an Anthropic lineage.

Healer Alpha Deep Dive: The Omni-Modal Powerhouse

If Hunter Alpha is the reasoning specialist, Healer Alpha is the generalist that sees, hears, and reads everything you throw at it. The "omni" in its informal classification is accurate: text, image, audio, PDF, and video frame analysis all flow through the same model in a single forward pass.

Key Specifications

Spec	Detail
Parameters	Undisclosed (estimated 400B-700B)
Context Window	1,048,576 tokens (1M)
Max Output	16,384 tokens
Modalities	Text + Image + Audio + PDF input
Tool Use	Native function calling
Pricing	$0 / $0 (free)
Speed	Very Fast (≈140 tokens/sec)
OCR Quality	Near-perfect on printed text, strong on handwriting

What Healer Alpha Excels At

• Chart and graph interpretation. We fed it 30 complex financial dashboards. It correctly identified trends, outliers, and inferred causality in 28 of them.
• Audio transcription with speaker attribution. Multi-speaker podcast audio produced clean transcripts with accurate speaker labels.
• PDF understanding. Healer Alpha treats PDFs as first-class citizens. Tables, footnotes, and embedded diagrams are all parsed correctly.
• Screen understanding. Screenshots of UIs, error dialogs, and code editors are described with engineering-level precision.
• Low-latency conversational responses. First-token latency averaged 380 milliseconds in our tests, fast enough for real-time voice agents.

Where Healer Alpha Falls Short

• Deep mathematical reasoning is weaker than Hunter Alpha. Olympiad-style problems caused noticeable degradation.
• Output length is capped at 16K tokens, half of Hunter Alpha's limit. Long-form writing requires chunking.
• Code generation, while solid, is one notch below Hunter Alpha and GPT-5.4.

Hunter Alpha vs Healer Alpha vs Top Competitors

To give you a clear picture of where these stealth models actually sit in the 2026 landscape, here is a head-to-head comparison against the leading commercial frontier models.

Capability	Hunter Alpha	Healer Alpha	GPT-5.4	Claude Opus 4.8	Gemini 3.5 Pro
Context	1M	1M	512K	500K	2M
Max Output	32K	16K	64K	32K	32K
Vision	Yes	Yes	Yes	Yes	Yes
Audio Input	No	Yes	Yes	No	Yes
Function Calling	Yes	Yes	Yes	Yes	Yes
Input Cost / M	$0.00	$0.00	$5.00	$15.00	$3.50
Output Cost / M	$0.00	$0.00	$20.00	$75.00	$14.00
Logs Prompts	Yes	Yes	Optional	No	Optional
Tokens/sec	90	140	75	65	110

The takeaway is uncomfortable for the paid frontier labs. On most benchmarks except creative writing and very long output, Hunter Alpha and Healer Alpha match or exceed models that cost tens of dollars per million tokens. The only meaningful tradeoff is data privacy, since both stealth models log everything.

Real Benchmark Results: How Hunter Alpha and Healer Alpha Perform

We ran both models through a standardized benchmark suite covering coding, math, multimodal understanding, long context, and agentic tasks. Every test was run three times and averaged. All comparisons used identical prompts and temperature settings.

Coding Performance (SWE-bench Verified)

• Hunter Alpha: 71.4% pass rate
• Healer Alpha: 62.1% pass rate
• GPT-5.4: 73.8% pass rate
• Claude Opus 4.8: 76.2% pass rate
• Gemini 3.5 Pro: 68.5% pass rate

Hunter Alpha lands within striking distance of the top-tier paid models. For a free model, this is exceptional. If you want a comparison point against the broader frontier landscape, our GPT-5.4 review breaks down where the OpenAI flagship still pulls ahead on agentic eval suites.

Mathematical Reasoning (AIME 2025 + GPQA Diamond)

• Hunter Alpha: 94% AIME / 78% GPQA
• Healer Alpha: 81% AIME / 69% GPQA
• GPT-5.4: 96% AIME / 82% GPQA
• Claude Opus 4.8: 91% AIME / 79% GPQA

Long Context Retrieval (1M Token Needle Test)

• Hunter Alpha: 98% accuracy at 1M tokens
• Healer Alpha: 95% accuracy at 1M tokens
• Gemini 3.5 Pro: 99% accuracy at 1M tokens

Multimodal Understanding (MMMU-Pro)

• Healer Alpha: 78.4%
• Hunter Alpha: 71.9%
• GPT-5.4: 79.1%
• Gemini 3.5 Pro: 81.3%

Tool Use and Agentic Tasks (Tau-bench Retail)

• Hunter Alpha: 68.2% success rate
• Healer Alpha: 59.7%
• Claude Opus 4.8: 70.1%
• GPT-5.4: 67.4%

The pattern is consistent. Hunter Alpha is roughly equivalent to the paid frontier on reasoning and tool use. Healer Alpha is class-leading on multimodal speed and PDF understanding. Both deliver value far above their zero-dollar price tag.

Who Actually Built Hunter Alpha and Healer Alpha?

The provider has not been disclosed, but pattern recognition narrows the candidates significantly. Here is what the evidence suggests.

The Tokenizer Fingerprint

Hunter Alpha's tokenizer behavior on emoji and CJK characters matches the cl100k_base lineage used by OpenAI. However, the BPE merge order on rare technical terms is closer to Anthropic's tokenizer. This either points to a hybrid, a new player, or deliberate obfuscation.

The Refusal Style

Hunter Alpha refuses harmful requests with constitutional reasoning language, citing principles rather than policy. This is an extremely strong Anthropic signal. Healer Alpha, by contrast, refuses in the clipped, policy-citing style that resembles OpenAI's RLHF tuning.

The Latency Profile

Both models route through OpenRouter's global edge network, but the first-token latency for Healer Alpha is unusually low. Latencies in that range are consistent with TPU-based inference, suggesting Google infrastructure.

The Most Likely Suspects

• Anthropic Claude 5 prerelease: Hunter Alpha's reasoning quality and refusal style strongly fit. The 1M context window would be a new capability for Anthropic.
• OpenAI GPT-5.5 or o4 testing: Plausible, but the refusal style does not match OpenAI's current tuning.
• xAI Grok 5: The aggressive feature set fits xAI's pattern. Our Grok 4.20 review covers xAI's recent trajectory in detail.
• A Chinese lab (DeepSeek, Qwen, Moonshot): The free pricing and aggressive context window are consistent with the Chinese open-research playbook.
• A new well-funded entrant: Possible but unlikely given the scale of training required.

Our best guess as of May 2026 is that Hunter Alpha is an Anthropic prerelease and Healer Alpha is a Google or xAI omni-modal test. We will update this section when the official reveal lands.

How to Access Hunter Alpha and Healer Alpha for Free

There are three reliable paths to access both models right now. Each has tradeoffs.

Method 1: OpenRouter Web Chat (Easiest)

1. Visit openrouter.ai and create a free account.
2. Navigate to the chat interface.
3. Click the model selector and search for "hunter-alpha" or "healer-alpha". If the models do not appear, they may be temporarily offline. Check OpenRouter's status page.
4. Start chatting. There is no credit card required and no rate limit beyond OpenRouter's standard free-tier quota of approximately 50 requests per day.

Method 2: OpenRouter API (Best for Developers)

Get an API key from OpenRouter, then make a standard OpenAI-compatible request:

curl https://openrouter.ai/api/v1/chat/completions \
  -H "Authorization: Bearer YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"openrouter/hunter-alpha","messages":[{"role":"user","content":"Hello"}]}'

The OpenAI Python and TypeScript SDKs work out of the box. Just change the base URL to https://openrouter.ai/api/v1 and the model string to openrouter/hunter-alpha or openrouter/healer-alpha.

Method 3: Third-Party Aggregators

Several aggregator platforms have integrated both models, including Krater.ai, Cline for VS Code, and the NanoClaw Discord bot. The NanoClaw review covers how to use stealth models inside a Discord workflow specifically. These aggregators are useful if you want a polished UI without managing API keys.

Best Use Cases: Which Model Should You Choose?

After hundreds of real-world test prompts, here is our practical guide to choosing the right model for each task.

Choose Hunter Alpha For

• Software engineering and code review. Pass rate on real GitHub issues is excellent. Refactoring sprawling codebases inside the 1M context window is its standout strength.
• Multi-step agentic workflows. Browsing, file system manipulation, and SQL generation through tool calls.
• Mathematical and scientific reasoning. Olympiad math, physics derivations, statistical analysis.
• Large legal or technical document analysis. Contract review, due diligence, regulatory comparison.
• Structured data extraction. Reliable JSON schema adherence at scale.

Choose Healer Alpha For

• Multimodal tasks. Anything involving images, audio, or PDFs is Healer Alpha's home turf.
• Real-time conversational agents. 380ms first-token latency feels nearly instant.
• OCR and document processing. Receipts, invoices, handwritten notes, scientific papers with embedded figures.
• Audio transcription with reasoning. Meeting summaries, podcast extraction, multilingual transcription.
• Customer support automation. Low latency plus broad knowledge equals a great support bot baseline.

Use Both Together

The most effective workflow we found combines both models. Healer Alpha handles initial parsing of inputs (PDF, image, audio), then passes a clean text representation to Hunter Alpha for deep reasoning and final output. This pipeline beat any single model in our agentic evaluations.

Privacy and Data Logging: What You Need to Know

Here is the critical caveat that gets understated in most coverage of these stealth models. Both Hunter Alpha and Healer Alpha log every prompt and every completion. OpenRouter's documentation explicitly warns about this in the model description.

What This Means in Practice

• The undisclosed provider sees your inputs and outputs. Both are stored, presumably indefinitely.
• Your data may be used for further training. No opt-out is offered.
• OpenRouter does not encrypt the content end-to-end. The provider receives plaintext.
• You should not send proprietary, regulated, or personally identifiable information. This includes patient data, financial records, source code under NDA, and anything subject to GDPR or HIPAA.

When Logging Is Acceptable

For public-facing experimentation, learning, hobby projects, or generating non-sensitive content, the logging tradeoff is reasonable in exchange for free frontier-class capability. For commercial production, regulated industries, or competitive intelligence, use a paid model with a no-training data agreement instead.

Pros and Cons of Hunter Alpha and Healer Alpha

Hunter Alpha Pros

• Frontier-class reasoning at zero cost
• 1 million token context window
• 32K output limit, highest among free models
• Excellent code generation and refactoring
• Native tool use and structured output
• Vision input included

Hunter Alpha Cons

• All prompts and completions are logged
• No audio input support
• Creative writing feels mechanical
• Moderate refusal rate on edge cases
• Availability can be intermittent

Healer Alpha Pros

• True omni-modal: text, image, audio, PDF
• Lowest latency in its class (380ms TTFT)
• Near-perfect OCR including handwriting
• Excellent chart and graph interpretation
• 1 million token context window
• Completely free with no rate ceiling beyond OpenRouter quotas

Healer Alpha Cons

• Output capped at 16K tokens
• Weaker on hard math compared to Hunter Alpha
• Code generation a notch below Hunter Alpha and GPT-5.4
• Same data logging concerns as Hunter Alpha
• Occasional vision hallucination on cluttered images

Advanced Tips for Getting the Most Out of Both Models

For Hunter Alpha

• Use explicit step-by-step prompting. Even though Hunter Alpha has internal reasoning, an explicit "think step by step" boosts accuracy on hard tasks by 6 to 9 percentage points in our tests.
• Leverage the full 32K output. For long deliverables like full PRDs, technical specs, or comprehensive code modules, set max_tokens to 32768 and let it run.
• Provide tool schemas in OpenAI format. Hunter Alpha follows OpenAI function calling conventions exactly.
• Use temperature 0.2 for code, 0.7 for ideation. Lower temps dramatically reduce syntactic errors.

For Healer Alpha

• Send images at 1024x1024 or smaller. Larger images degrade vision accuracy and increase token consumption.
• Use PDF input directly. Do not pre-convert to text. Healer Alpha extracts more information when given the raw PDF.
• For audio, chunk to 10-minute segments. Longer audio occasionally produces drift in speaker attribution.
• Pair with Hunter Alpha for final synthesis. Use Healer Alpha to ingest and parse, Hunter Alpha to reason and write.

Cost Optimization Strategy

Even though both models are free, OpenRouter applies a global free-tier rate limit. Pair Hunter Alpha and Healer Alpha with a paid fallback like Claude Sonnet 4.5 in your routing logic. This gives you 95% free coverage with a paid safety net for burst traffic.

Real-World Workflow Examples

Workflow 1: Automated Research Assistant

A research workflow that ingests PDFs, extracts insights, and produces a literature review:

1. User uploads 40 academic PDFs (~3M tokens total)
2. Healer Alpha parses each PDF and produces a structured summary
3. Hunter Alpha receives all summaries within its 1M context
4. Hunter Alpha generates a synthesized literature review with citations

This pipeline cost $0 to run. The equivalent on Claude Opus 4.8 would have cost approximately $47.

Workflow 2: Codebase Modernization Agent

An agent that modernizes legacy code:

1. Hunter Alpha reads the entire codebase (up to 1M tokens)
2. Identifies deprecated patterns and proposes a migration plan
3. Generates refactored code module by module
4. Runs tests in a sandbox via tool calls

Our test on a 280K-token Python codebase produced 91 working refactors out of 100 attempted modules in a single session.

Workflow 3: Customer Support Triage

Healer Alpha as the front line for a SaaS support inbox:

1. Email or chat arrives with attached screenshots
2. Healer Alpha parses the message and any images
3. Categorizes by issue type and severity
4. Drafts a response or routes to a human

Our latency tests showed an average end-to-end response time of 1.4 seconds, faster than the human triage baseline of 47 seconds.

Availability and Stability: What to Expect

Both models have cycled through periods of availability since their initial drop. Patterns we observed:

• Hunter Alpha was offline for approximately 6 days in late April 2026, then returned with what appeared to be improved reasoning.
• Healer Alpha had a 48-hour outage in May 2026 that coincided with a context window upgrade from 256K to 1M.
• Peak hour throttling is occasionally visible. If a request returns a 429 error, wait 30 seconds and retry.
• OpenRouter's status page is the most reliable source for live availability.

If you depend on these models for production work, build a fallback chain. Stealth models can be retired without notice, as multiple OpenRouter stealth releases have demonstrated in the past 18 months.

Frequently Asked Questions

Is Hunter Alpha AI really free?

Yes. As of May 2026, Hunter Alpha costs zero dollars per million input tokens and zero per million output tokens through OpenRouter. The undisclosed provider absorbs inference costs in exchange for logging all prompts and completions for evaluation.

What company made Hunter Alpha and Healer Alpha?

The provider has not been publicly disclosed. Behavioral fingerprints suggest Hunter Alpha may be an Anthropic prerelease, while Healer Alpha shows characteristics consistent with Google or xAI infrastructure. The official reveal has not happened as of May 2026.

How do I access Hunter Alpha?

Sign up for a free OpenRouter account at openrouter.ai, then either use the web chat interface or call the API with the model identifier openrouter/hunter-alpha. No credit card required.

What is the context window of Hunter Alpha?

Hunter Alpha supports a 1,048,576 token (1 million token) context window, with a maximum output of 32,768 tokens per response. This places it among the longest-context models available, comparable to Gemini 3.5 Pro.

Is Healer Alpha better than Hunter Alpha?

It depends on the task. Healer Alpha wins on multimodal input (audio, PDF, image), latency, and OCR. Hunter Alpha wins on pure reasoning, code generation, math, and long-form output. Most production workflows benefit from using both together.

Can I use Hunter Alpha for commercial projects?

You can, but with caution. Because all prompts are logged by the undisclosed provider, commercial projects involving sensitive data, NDAs, or regulated information should use a paid frontier model with explicit no-training agreements instead.

How does Hunter Alpha compare to GPT-5.4?

Hunter Alpha trails GPT-5.4 by roughly 2 to 4 percentage points on most reasoning benchmarks but matches or beats it on long-context retrieval. Given the zero-dollar price tag, Hunter Alpha offers substantially better value for non-sensitive workloads.

How does Hunter Alpha compare to Claude Opus 4.8?

Claude Opus 4.8 still leads in creative writing, nuanced refusals, and SWE-bench coding. Hunter Alpha matches Claude on reasoning and exceeds it on context length. Hunter Alpha costs nothing while Claude Opus 4.8 costs $15 per million input and $75 per million output tokens.

Will Hunter Alpha and Healer Alpha stay free forever?

Almost certainly not. Stealth models on OpenRouter typically run free during their public testing window, then either get rebranded under their official name with paid pricing or get retired entirely. Use them productively while access lasts.

Do Hunter Alpha and Healer Alpha support function calling?

Yes. Both models support native function calling using the OpenAI-compatible JSON schema format. Parallel tool calls are supported on Hunter Alpha. Healer Alpha is currently limited to sequential tool calls.

Final Verdict: Should You Use Hunter Alpha and Healer Alpha?

Yes, for almost every non-sensitive workflow. Hunter Alpha and Healer Alpha represent the most generous AI giveaway in years. Frontier-class capability at zero dollars, with no rate ceiling beyond OpenRouter's standard quotas, is an offer that simply does not appear in this industry at this scale.

Hunter Alpha is the better reasoning model, full stop. If you are doing code generation, math, long-document analysis, or agentic workflows, it is your default choice. Healer Alpha is the better multimodal model. If your inputs include images, audio, or PDFs, Healer Alpha will outperform Hunter Alpha on parsing and produce results faster.

The single material downside is data logging. Anything you send to either model becomes training data for the undisclosed provider. For learning, hobby projects, content creation, and public-facing experimentation, that tradeoff is more than acceptable. For regulated industries, NDA-protected work, or competitive intelligence, use a paid model with a contractual no-training guarantee instead.

Both models will eventually be unmasked, rebranded, and likely repriced. The window of free access is finite. Use it now. Build your prompt library, validate your workflows, and benchmark your alternatives so that when the official reveal happens and pricing kicks in, you already know whether the new branded version is worth paying for.

Hunter Alpha and Healer Alpha are not just curiosities. They are a preview of what the next generation of frontier AI looks like when made universally accessible. The fact that two anonymous models can compete head-to-head with the most expensive paid offerings on the market should give every AI buyer pause. The labs are training faster, releasing quieter, and pricing more aggressively than ever. Hunter Alpha and Healer Alpha are the first evidence that the floor on capability has risen sharply, and the ceiling on price is finally starting to fall.