GPT-6 Astra, Claude Fable, Gemini 3.8: A Busy Week for New LLM Releases
Weekly digest from LLM Price Watch — tracking live pricing across OpenAI, Anthropic, Google, DeepSeek, and xAI.
No price changes this week across any of the five providers we track. Zero. But the model catalog? That got noticeably busier. Between September 2nd and 5th, we spotted nine new model IDs across four providers. Here's what showed up and what's worth paying attention to.
OpenAI: GPT-6 Astra (and a Pro tier)
The biggest names to land on our radar are openai/gpt-6-astra and openai/gpt-6-astra-pro, both spotted on September 5th. Both also have batch variants (gpt-6-astra:batch and gpt-6-astra-pro:batch).
The GPT-6 naming is a notable jump — we've been sitting with GPT-4 variants for a long time in practical use. The "Astra" label has been associated with Google's multimodal work in the past, so it's an interesting name choice from OpenAI. The Pro tier suggests a familiar tiered pricing structure is coming, similar to what we've seen with other model families.
Practically speaking: if you're building something today, I'd hold off on committing to these until pricing is confirmed and benchmarks are out. Batch variants being available at launch is a good sign — batch processing has been one of the more reliable ways to cut costs on longer, async workloads.
xAI: Grok 4.3 Batch
x-ai/grok-4.3:batch appeared on September 4th. This is a batch-only variant, no standard inference version detected yet. xAI has been iterating fairly quickly on the Grok 4 line, and adding batch support makes sense if they're targeting workloads where latency isn't critical — think evals, document processing, data pipelines.
Google: Gemini 3.8 Flash
google/gemini-3.8-flash and its batch counterpart showed up on September 3rd. The Flash line from Google has generally been their cost-efficient, faster-response tier — good for high-volume use cases where you don't need the full Gemini Pro/Ultra capability. Gemini 3.8 continuing that Flash tradition makes sense. If past Flash pricing holds, this could be one of the cheaper options in the new generation lineup once pricing is confirmed.
Anthropic: Claude Fable 5.1
anthropic/claude-fable-5.1 and anthropic/claude-fable-5.1:batch appeared on September 2nd, making it the earliest arrival in this week's batch.
The "Fable" name is new for Anthropic — we haven't tracked a Fable line before. It's not clear yet where this sits relative to Sonnet, Haiku, or Opus in terms of capability or price point. The name might suggest a model tuned for narrative or creative tasks, but that's speculation. Worth watching for the official positioning.
What This Week Actually Means
With no price changes, your current cost modeling stays the same. Nothing got cheaper or more expensive in the existing lineup.
The main thing to flag is that four providers dropped new model IDs in a four-day window. That's a lot of surface area appearing at once. A few practical notes:
- Don't route production traffic to these yet unless you've tested them. New model IDs appearing in the API doesn't always mean they're fully stable or that pricing is final.
- Batch variants at launch is genuinely useful — it signals the providers are thinking about cost-sensitive workloads from day one.
- The GPT-6 naming is the most significant signal here. If OpenAI is moving to a 6.x generation label, expect Anthropic and Google to respond with their own positioning in the coming weeks.
I'll be watching for actual pricing data on all of these. Once numbers are confirmed, I'll do a proper comparison.
For live pricing across all tracked models and providers, check llmpricewatch.com.
Top comments (0)