Key takeaways
- OpenRouter dates the new inclusionai/ling-3.2-flash:free listing to November 13, 2026. This establishes the registry listing date, not a separately verified developer announcement date.
- The registry lists text and image input, text output, 1,048,576 context tokens and a top-provider maximum completion of 262,144 tokens. The page advertises up to 200 images per request and optional thinking.
- Input and output token prices are both zero at retrieval. Free pricing does not establish unlimited requests, guaranteed capacity or a permanent commercial offer.
- The linked model card returned HTTP 401, and no API key was available for testing. We have not verified downloadable weights, a license, benchmark quality or real-world image and long-context performance.
What is available now
OpenRouter lists inclusionAI’s Ling 3.2 Flash as a free model released on November 13, 2026. Builders can select inclusionai/ling-3.2-flash:free through OpenRouter. The page describes a hybrid reasoning model that accepts text and images and produces text. This is news about a currently listed API option, rather than a claim that AccessAllGPT has confirmed a separate developer launch announcement.
At our November 14 retrieval, the public registry reports a 1,048,576-token context window and a top-provider maximum completion of 262,144 tokens. These are advertised service limits, not measurements of how reliably the model uses a million tokens or whether every request can fill both limits simultaneously. The model page additionally advertises up to 200 images per request and says thinking can be switched off per request.
The listing gives input and output token prices of $0 per million tokens. That makes the model worth considering for a small evaluation without token charges at the listed rates. It does not promise unlimited traffic, reserved capacity, a service-level agreement or unchanged pricing in the future. Check current account limits and provider availability before building a dependency on the free route.
Why this is useful to builders
The combination matters more than any single headline specification: a free route, image input, a large advertised context and optional reasoning lets a team compare document and screenshot workflows against its existing model without starting with a paid inference commitment. That is an evaluation opportunity, not evidence that Ling is a replacement for a stronger or more reliable model.
OpenRouter’s registry lists tools, tool_choice, response_format, structured_outputs and reasoning among supported parameters. Treat that as advertised interface support. It does not establish schema adherence, correct tool selection or safe execution. A model can accept a parameter and still produce an incorrect answer or an unusable tool call.
Our analysis: begin with short, checkable tasks before paying the latency and review cost of a very large context. A small screenshot-reading task can reveal whether details are grounded in the image; a fixed document question with known answers can reveal whether the model actually retrieves the relevant passage. Neither requires uploading confidential material.
What remains unverified
The OpenRouter page links to inclusionAI/Ling-3.2-flash on Hugging Face as “Model weights.” That page and its model API returned HTTP 401 during this run, while the raw README returned an authentication error. We could not establish whether this reflects access restrictions, an unavailable repository or another publication issue. A weights link alone is not proof of a publicly downloadable release.
We therefore make no claim about the new model’s license, parameter count, architecture, training data, downloadable artifacts or self-hosting requirements. Older Ling releases cannot supply those facts for this model. Readers evaluating local deployment should wait for an accessible model card, license and actual artifacts.
No OPENROUTER_API_KEY was available in this publishing environment. AccessAllGPT did not call the model, run the UI-generation harness, inspect model-generated screenshots or measure speed, accuracy, tool use or long-context recall. The advertised 200-image allowance and switchable reasoning are service descriptions, not results we reproduced. No benchmark comparison is established by this briefing.
What to do next
If you already use OpenRouter, add the exact free model ID to a small, non-sensitive comparison set. Use the same prompts and expected answers as your current model, record failures as well as successes, and compare reasoning-on and reasoning-off behavior only after checking the current request documentation. Inspect the provider’s data-handling terms before sending anything private.
Keep a working fallback and confirm that your application distinguishes a failed or unavailable free request from a valid empty answer. For buyers who need licensed downloadable weights, an uptime commitment or proven task quality, the next step is to wait for those specific missing pieces rather than infer them from a zero price or a large context window.
A first evaluation, not a production endorsement
This is a proposed checklist; AccessAllGPT did not execute it in this run.
Entries stay in this browser tab and are not submitted to AccessAllGPT. Blank responses are copied as [Unresolved].
Confirm inclusionai/ling-3.2-flash:free is available to your account, along with current limits and provider terms.
Use public or synthetic documents and images with known answers. Record omissions, invented details and invalid structured output.
Compare the documented thinking settings on the same tasks; record correctness and latency rather than assuming more thinking is better.
Keep an API fallback. Require an accessible license and actual weights before treating this as a self-hosting option.
Primary sources
Browse the publication-wide evidence index →
- Ling 3.2 Flash (free): model pageOpenRouter · Reviewed: Model overview; November 13 release date; modalities; price; context; provider listing; linked model weights · Retrieved · Supports: Displayed listing date, exact model ID, hybrid reasoning description, up to 200 images, optional thinking and advertised free API access. These are OpenRouter descriptions, not independently reproduced capabilities.
- OpenRouter public model registryOpenRouter · Reviewed: inclusionai/ling-3.2-flash:free record: architecture, context_length, pricing, top_provider and supported_parameters · Retrieved · Supports: Exact registry ID, text/image input and text output, 1,048,576-token context, 262,144 maximum completion tokens, zero prompt/completion prices and advertised tools, structured output and reasoning parameter support.
Limitations
Registry-backed primary-source briefing, not a hands-on test. No OpenRouter API key was set and no inference request was made. The linked developer model card and API returned HTTP 401, so license, downloadable weights, architecture and developer release chronology remain unverified. OpenRouter’s November 13 date is a listing date. Limits, pricing and routing can change; no benchmark, unlimited capacity or production reliability is established.
Disclosures
AccessAllGPT received no payment, credits or private access for this briefing. Neither inclusionAI nor OpenRouter reviewed or endorsed it. AccessAllGPT Research is operated by NeuralArc, is independent and is not affiliated with inclusionAI or OpenRouter. Publication-wide relationships are listed on the disclosures page.
