New Models Only This Week: Gemini 4, Claude Fable, and Kimi Show Up in the Tracker
No price changes this week. Zero. If you're watching the tracker hoping for a cost drop to justify a migration, nothing moved. But the new model list is genuinely interesting, so let's talk about that instead.
Here's what our tracker picked up between November 10–13, 2026:
Google dropped two Gemini 4 variants
Gemini 4 Carbon appeared on November 11, alongside a batch variant. Two days later, Gemini 4 Hydrogen showed up, also with a batch option.
The naming convention here — Carbon, Hydrogen — suggests Google is moving toward element-based tiering within the Gemini 4 family, similar to how they've used "Flash" and "Pro" historically. Hydrogen is lighter than Carbon on the periodic table, which probably maps to a faster/cheaper vs. more capable distinction, though that's inference on my part. The actual pricing and capability specs will tell the real story once they're more widely documented.
The batch variants for both are worth noting. If you're running offline workloads — evaluations, document processing, anything that doesn't need real-time responses — batch endpoints typically come at a meaningful discount. Worth checking the pricing on those specifically before defaulting to the standard endpoint.
Anthropic added Claude Fable 5.2
Claude Fable 5.2 appeared on November 10, with a batch variant also available from day one.
The "Fable" naming isn't part of Anthropic's traditional Claude lineup (Haiku, Sonnet, Opus), which makes this one to watch. It could be a specialized model — creative writing, instruction-following, something task-specific — or it could signal a new product line entirely. The 5.2 version number suggests there was at least a 5.0 and 5.1 before this, though those may have been internal or limited-access releases.
If you're building anything on Anthropic's API, it's worth pulling the model metadata to see how Fable 5.2 is positioned relative to the Sonnet/Opus tier you're currently using.
Moonshot AI's Kimi K3.1 Pro got a batch variant
Kimi K3.1 Pro batch showed up on November 12. This isn't a new model per se — K3.1 Pro has been around — but the batch endpoint being available now matters if you've been using this model for high-volume work and paying standard endpoint prices.
Moonshot AI has been consistently competitive on price for longer-context tasks, and batch pricing usually makes that gap wider.
Qwen's batch expansion continues
Qwen3.8-max-0902 batch was spotted on November 10. Again, this is an existing model getting a batch endpoint added. Qwen has been expanding batch availability across its lineup fairly steadily, which is useful if you're doing large-scale inference on a budget.
What this week actually means for you
If you're in "actively choosing a model" mode right now:
- No existing prices changed, so your current cost estimates are still valid.
- Two new Gemini 4 variants are worth evaluating if you're in the Google ecosystem and want to see where the new tier lands quality-wise.
- Claude Fable 5.2 is an unknown quantity with an unusual name — if you're Anthropic-dependent, it's worth running your eval suite against it.
- Batch endpoints got added to three models this week. If you're not using batch for async workloads, you're probably leaving money on the table.
Nothing here is urgent. It's more of a "update your model shortlist and check pricing before your next architecture decision" kind of week.
I track this every week at LLM Price Watch, where you can see live pricing across Claude, GPT, Gemini, DeepSeek, and Grok. If you want an alert when something actually changes, the site has that too.
Top comments (0)