GPT-6 Astra, Claude Fable, Gemini 3.8: A Busy Week for New LLM Releases
Weekly digest from LLM Price Watch — tracking real pricing data across OpenAI, Anthropic, Google, DeepSeek, and xAI.
No price changes to report this week across our tracked providers. But the model catalog grew significantly — nine new model IDs showed up in our tracker between September 2nd and 5th, 2026, spanning all four of the big players we watch. Here's what appeared and what's worth noting practically.
OpenAI: GPT-6 Astra (and a Pro tier)
The biggest news by name alone: OpenAI dropped two new model families on September 5th — gpt-6-astra and gpt-6-astra-pro, each with a corresponding :batch variant.
The batch variants are the immediate practical signal here. OpenAI has consistently priced batch inference at a meaningful discount (typically around 50% off standard rates), so if you're running non-latency-sensitive workloads — evaluations, document processing, bulk classification — the batch endpoints are worth watching closely once pricing is confirmed.
The split between a base gpt-6-astra and gpt-6-astra-pro suggests a tiered capability structure, similar to how the GPT-4o / GPT-4o-mini split played out. Whether the Pro variant is meaningfully better for reasoning-heavy tasks or just a marketing distinction remains to be seen. I'd hold off on routing production traffic until pricing and benchmark data are clearer.
xAI: Grok 4.3 Batch
On September 4th, grok-4.3:batch appeared — notably, just the batch variant, with no corresponding standard model showing up in our tracker this week. That's an interesting release pattern. It could mean the base grok-4.3 was already present in the catalog, or that xAI is leading with batch access before a broader rollout. Either way, if you're already using Grok models for async workloads, this is worth checking against your current setup.
Google: Gemini 3.8 Flash
September 3rd brought gemini-3.8-flash and its batch counterpart. The Flash line from Google has generally been their price-performance sweet spot — faster and cheaper than Pro variants, with quality that's often sufficient for structured output tasks, summarization, and classification.
If the 3.8 Flash follows the same pricing trajectory as earlier Flash models, it could become a solid default for high-volume, cost-sensitive applications. The batch variant showing up simultaneously with the standard model is a good sign — it suggests Google is treating batch access as a first-class feature rather than an afterthought.
Anthropic: Claude Fable 5.1
The most intriguing naming this week: claude-fable-5.1 appeared on September 2nd, again with a batch variant. "Fable" isn't a naming convention we've seen from Anthropic before — their previous lines have been Haiku, Sonnet, and Opus. A new name likely means a new positioning, possibly a specialized or fine-tuned variant rather than a general-purpose tier.
Without more documentation from Anthropic, it's hard to know where Fable sits in their lineup — whether it's meant to replace something, complement the existing tiers, or target a specific use case. The 5.1 version number suggests it's not a ground-up new model but an iteration of something in the Claude 5 family.
What This Means Practically
A week with nine new model IDs and zero price changes is a week for watching, not necessarily acting. A few things I'd suggest:
- Don't rush to migrate to any of these until pricing is confirmed and published. New model IDs on OpenRouter don't always come with finalized pricing on day one.
- Batch variants are worth bookmarking across all four providers. If your workload tolerates async processing, batch endpoints typically offer the best cost efficiency available.
- The Anthropic "Fable" naming is unusual enough to keep an eye on — it may signal a meaningful capability shift or specialization that affects how you'd evaluate it against Sonnet or Haiku.
I'll be tracking pricing as it settles across all nine of these models in next week's digest.
For live pricing and model tracking across Claude, GPT, Gemini, DeepSeek, and Grok, visit llmpricewatch.com.
Top comments (0)