JUNE 13-14 2026 / DATE-HONEST LIST / THE ROUTING TRAP

A fresh open model is easy to download. The hard part is making your agent loop actually send it a request.

Every roundup of these two days hands you model names and a column of context-window specs. None of them cover the step after that: once you have a model worth testing, wiring it into a desktop agent loop is a base-URL override, and that override quietly routes only one family of models. This page gives the date-honest answer first, then the trap that sends a brand-new model exactly zero requests.

Matthew Diakonov, Written with AI

Published June 17, 20268 min read

Direct answer, verified 2026-06-17

No major model carries a precise June 13 or June 14, 2026 release date. The nearest dated drops bracket the window rather than landing on it, per llm-stats.com/ai-news:

Claude Fable 5 (Anthropic) on June 9.
DiffusionGemma 26B-A4B (Google) on June 10.
GLM-5.2 (Zhipu AI) around June 17.

Because no platform publishes a list keyed to a calendar day, the only date-honest sources for any 48-hour window are huggingface.co/models sorted by trending, huggingface.co/papers, and github.com/trending.

That is the literal answer. The rest of this page is the part the spec tables skip: getting one of those models into your agent loop, and the override that silently routes only Claude-family traffic.

Two ways a fresh model reaches the loop

The Claude-family drop in this window (Fable 5, dated June 9) is the easy case: a client that already speaks the Anthropic API talks to it with no extra setup. The interesting case is an open-weight model like DiffusionGemma or a post-launch MiniMax M3 weight drop. To run one of those in the same loop, you put a bridge in front of it that exposes the Anthropic message format, then point the client at the bridge.

In Fazm that pointer is the custom API endpoint setting, and under the hood it overrides one thing: the ANTHROPIC_BASE_URL environment variable. That single fact is the whole story, because the override is scoped to traffic that travels the Anthropic path. Models that travel a different path never see it.

Where the request actually goes

A single model-id classification decides everything. If the id is Claude-routed, the request rides the Anthropic path and your override applies. If the id is a Gemini or Codex model, it travels its own backend and the bridge you so carefully stood up receives nothing.

Custom endpoint routing, by model family

Nothing throws in the bypass case. The session runs, an answer comes back, and it simply came from a model that is not the one you were trying to test. That is what makes it a trap rather than a visible error: the default selection for a new account is Gemini Flash, so the common path is the one that skips your bridge.

The check that landed June 15: isClaudeModelId, by exclusion

Two commits on June 15, 2026 turn the trap into a visible warning. First, a helper that names the routing rule the client was already relying on. It is defined by exclusion on purpose: a model is Claude-routed if it is neither a Gemini model nor a Codex model, which keeps the client in lockstep with the bridge's own regexes and means a new Claude id works the instant it exists.

Desktop/Sources/Providers/ChatProvider.swift (commit 473b8a68)

/// True if the model id routes through the Anthropic SDK path (Claude).
/// Defined by exclusion: anything that is neither a Gemini nor a Codex
/// model goes through the Anthropic path. This matters for the Custom
/// API Endpoint setting, which only overrides `ANTHROPIC_BASE_URL` and
/// therefore ONLY affects Claude-routed traffic - Gemini/Codex models
/// bypass it entirely.
static func isClaudeModelId(_ modelId: String) -> Bool {
    return !isGeminiModelId(modelId) && !isCodexModelId(modelId)
}

Second, the settings panel uses that helper to guard a warning. When the custom endpoint is on and the selected model is not Claude-routed, an orange box appears with a one-click button to switch. The inline comment is blunt about why it exists: without it, the endpoint silently receives zero requests.

Desktop/Sources/MainWindow/Pages/SettingsPage.swift (commit ead89cf8)

// The custom endpoint only overrides ANTHROPIC_BASE_URL, so it
// ONLY routes Claude-model traffic. Gemini/Codex models bypass it
// entirely. New users default to Gemini Flash, so without this
// warning the endpoint silently receives zero requests.
if !ChatProvider.isClaudeModelId(shortcutSettings.selectedModel) {
    // orange warning + "Switch to a Claude model" button
}

The detail that makes this checkable rather than marketing: the warning is not a generic banner. It is gated by the exact predicate that governs routing, so it shows up precisely when, and only when, your current selection would skip the bridge.

Wiring a June drop into the loop without falling in

The order of operations is what keeps the bridge from going silent. Step three is the one the trap hides, and the warning from June 15 exists to make step four honest.

From open weights to a request your bridge sees

1
Stand up an Anthropic-format bridge
Serve the fresh open-weight model behind an endpoint that speaks the Anthropic API. A raw Gemini or OpenAI key will not work here; the bridge has to translate to and from the Anthropic message shape.
2
Enable the custom endpoint
In settings, turn on the custom API endpoint and paste your bridge URL. This overrides ANTHROPIC_BASE_URL for the app, and Fazm stops sending its built-in Anthropic key for that traffic.
3
Select a Claude-routed model
This is the step the trap hides. The override only touches Claude-routed traffic, so the selected model must be a Claude id. A Gemini or Codex selection bypasses the bridge and it receives nothing.
4
Confirm requests are landing
Run one turn and watch your bridge log. If the orange warning is showing in settings, your selection is not Claude-routed and the bridge will stay silent until you switch.

Why this belongs next to a release list

A roundup that stops at model names answers half the question. The half that decides whether the news is useful to you is whether you can get the new model to answer a real task in the loop you already use. For Claude-family drops that is trivial. For open-weight drops it is a bridge plus one routing rule, and the rule is the kind of thing you only learn after a wasted session unless the client tells you up front.

That is the whole reason a thirty-line settings warning earns a place beside a model launch. The releases are loud and frequent; the thing that decides whether you can actually evaluate one is quiet and lives in your client. On June 15 it became a warning you cannot miss.

Verify every claim here

For the release dates: open llm-stats.com/ai-news and check that Claude Fable 5 is dated June 9, DiffusionGemma 26B-A4B June 10, and GLM-5.2 mid-June, with nothing major pinned to June 13 or 14.
For the live landscape: open huggingface.co/papers and github.com/trending. The trending order shifts hour to hour, but the submission and commit dates are stable.
For the Fazm change: open github.com/mediar-ai/fazm, read Desktop/Sources/Providers/ChatProvider.swift for the isClaudeModelId helper and Desktop/Sources/MainWindow/Pages/SettingsPage.swift for the warning, then check the git log for commits 473b8a68 and ead89cf8, both authored June 15, 2026.

Want to test a fresh open-weight model in a real Mac agent loop?

Fifteen minutes. I will wire a model behind an Anthropic-format bridge, point a custom endpoint at it, and show the warning fire when the selected model would skip the bridge.

Frequently asked questions

What AI model released on June 13 or June 14, 2026?

Nothing major is dated precisely to either day. The public release trackers show the nearest dated drops bracketing the window rather than landing on it: Claude Fable 5 from Anthropic on June 9, DiffusionGemma 26B-A4B from Google on June 10, and GLM-5.2 from Zhipu AI around June 17. The reason a calendar-day query rarely returns a clean answer is that neither Hugging Face nor GitHub publishes a list keyed to a date. Both order discovery by a rolling trending score with no notion of when something shipped, so the date-honest feeds for any 48-hour window are huggingface.co/models sorted by trending, huggingface.co/papers, and github.com/trending.

Was this a quiet stretch for open weights?

No. The first half of June 2026 was dense. MiniMax M3 landed June 1 as an open-weight coding model with a one-million-token context window, with weights and a technical report promised shortly after launch. NVIDIA shipped Nemotron 3 Ultra on June 4, Google posted DiffusionGemma on June 10, and Zhipu's GLM-5.2 followed mid-month. The calendar simply does not pin a flagship to June 13 or 14 specifically. A vendor's own benchmark numbers are self-reported until an independent harness re-runs them, so treat any fresh drop as a thing to test, not a leaderboard fact.

Once a fresh open-weight model exists, how do I actually run it in an agent loop on a Mac?

Two paths. If it is a Claude-family model (Fable 5 and the rest of the Claude line), a wrapper that speaks the Anthropic API talks to it directly with no extra setup. If it is an open-weight or third-party model, you stand up a bridge that exposes it in the Anthropic API format (a local LLM bridge, a corporate proxy, or a GitHub Copilot bridge) and point your client at that bridge with a custom endpoint. In Fazm that custom endpoint setting overrides the ANTHROPIC_BASE_URL environment variable, which is what reroutes the traffic.

What is the routing trap this page is about?

A custom endpoint that overrides ANTHROPIC_BASE_URL only reroutes Claude-routed traffic. Gemini-family and Codex-family models do not go through the Anthropic path, so they bypass the override entirely. If your selected model is not a Claude id and you point the endpoint at a bridge serving a brand-new open model, the bridge receives zero requests and nothing visibly errors. New Fazm users default to Gemini Flash, which makes the silent-zero-requests case the common one, not an edge case.

How does Fazm catch the silent-zero-requests case?

On June 15, 2026 Fazm added a check in Desktop/Sources/MainWindow/Pages/SettingsPage.swift (commit ead89cf8). When the custom endpoint is enabled and the selected model is not a Claude id, the settings panel shows an orange warning that reads, in effect, your current model does not use this endpoint, the custom endpoint only applies to Claude models, plus a one-click button to switch to a Claude model. The check calls ChatProvider.isClaudeModelId, a helper added the same day (commit 473b8a68) that is defined by exclusion: a model is Claude-routed if it is neither a Gemini model nor a Codex model.

Why is isClaudeModelId defined by exclusion rather than a prefix list?

Because the routing decision already lived in two positive checks. The bridge classifies a model as Gemini (gemini-*, auto-gemini-*) or as Codex (gpt-*, codex-*, o[0-9]-*); anything that matches neither falls through to the Anthropic SDK path. Defining isClaudeModelId as not-Gemini-and-not-Codex keeps the client in lockstep with the bridge's own regexes, so a new Claude model id works the moment it exists without anyone updating a hardcoded allowlist. The in-source comment on the helper states exactly this and notes that the custom endpoint only affects Claude-routed traffic.

Does this mean I cannot run a fresh open-weight model in Fazm at all?

You can, but you route it deliberately. The clean path is to expose the open-weight model through an Anthropic-API-compatible bridge and keep your selected model on the Claude path so the ANTHROPIC_BASE_URL override actually applies. The warning exists so you find out before a session, not after, that a Gemini or Codex selection would have skipped your bridge. Fazm also will not send its built-in Anthropic key or count custom-endpoint usage against built-in credits, so a bridge run stays on your own keys.

How do I verify the Fazm change myself?

Open github.com/mediar-ai/fazm. Read Desktop/Sources/MainWindow/Pages/SettingsPage.swift for the orange warning block guarded by an isClaudeModelId check, and Desktop/Sources/Providers/ChatProvider.swift for the isClaudeModelId helper and its comment about the ANTHROPIC_BASE_URL override. The git log shows commit ead89cf8, 'Add warning for non-Claude models in custom endpoint settings,' and commit 473b8a68, 'Add isClaudeModelId helper and expose isCodexModelId,' both authored June 15, 2026. Everything is public, so none of this rests on faith.

Adjacent reading