GPT-6 Astra, Claude Fable, Gemini 3.8: A Busy Week for New LLM Releases
Weekly LLM pricing digest — week of September 5, 2026
No price changes to report this week across the five providers we track (OpenAI, Anthropic, Google, DeepSeek, and xAI). But the model catalog grew noticeably in just a few days, so let's walk through what actually showed up and what it might mean if you're picking a model right now.
OpenAI: GPT-6 Astra lands with a Pro tier and batch variants
The biggest news this week is that OpenAI dropped two new models on September 5th: gpt-6-astra and gpt-6-astra-pro, each with a corresponding :batch variant.
The naming convention here is familiar — the :batch versions are typically lower-cost, asynchronous endpoints suited for workloads where you don't need a real-time response. If you're doing large-scale document processing, evals, or anything that can tolerate a delay, batch is almost always the right call financially.
The split between a base astra and an astra-pro tier suggests OpenAI is continuing the pattern of offering a more capable (and presumably more expensive) variant alongside a leaner everyday option. Worth checking current pricing before assuming which one fits your use case — the gap between base and pro tiers has varied a lot across model generations.
Anthropic: Claude Fable 5.1
Anthropic's new entry, claude-fable-5.1, appeared on September 2nd, again with a batch variant.
The "Fable" name is new in Anthropic's lineup. Whether this sits above, below, or alongside the Sonnet/Opus/Haiku naming structure isn't something our tracker can tell you — we detect models as they appear on OpenRouter, not their internal capability tiers. What we do know is it showed up with batch support from day one, which suggests Anthropic is treating batch as a first-class offering rather than an afterthought.
If you're currently using Claude for any high-volume async workload, it's worth benchmarking fable-5.1 against whatever you're running now.
Google: Gemini 3.8 Flash
gemini-3.8-flash arrived on September 3rd, with batch support included.
The Flash line has consistently been Google's price-performance workhorse — fast, cheap, and good enough for a wide range of tasks. 3.8 continuing that series is a reasonable assumption, though again, actual pricing on our tracker is what should drive your decision, not the name alone.
If you're already using Gemini 2.x Flash for something latency-sensitive or cost-sensitive, this is probably worth a quick test.
xAI: Grok 4.3 Batch
xAI added grok-4.3:batch on September 4th. This appears to be a batch variant of an existing model rather than a completely new release, which fits the pattern of providers progressively rolling out async options for their model lineup.
What this week means practically
Five new model IDs in four days across four different providers, with zero price changes on existing models. A few things stand out:
Batch variants are becoming standard. Every new model this week came with a batch option. If you're not already routing eligible workloads to batch endpoints, you're likely leaving money on the table.
The catalog is expanding faster than pricing is changing. That means the cost landscape for existing models is stable right now, but the choice of which model to use is getting more complex. It's worth revisiting your model selection if you haven't done so recently.
None of these models have a pricing history yet. First-week pricing on a new model sometimes gets revised. It's not unusual to see adjustments in the first few weeks after launch.
We'll be tracking all of these as they settle in. If prices shift on any of the new models — or on anything else across our five providers — it'll show up in next week's digest.
For live pricing on all of these models and more, check LLM Price Watch.
Top comments (0)