What Together AI announced
On September 13, 2026, the @togethercompute account on X said that DeepSeek V4.1 Flash is now available on Together AI. The post lists the model's specifications: 552 billion total parameters, a 1M-token context window, and native multimodal input. That is the entire public basis for the release in the material reviewed — a single post, with no linked model card or pricing page.
The claim against GPT-5.6 Sol
The headline claim is comparative. According to the post, DeepSeek V4.1 Flash outperforms GPT-5.6 Sol on agentic benchmarks while costing one third per task. If that holds, the combination matters: better agentic scores at a third of the cost is what would move teams picking a model for tool-using workloads, because agent runs multiply spend by the number of steps taken. The post does not name the benchmarks, give scores, or specify what the "one third" is measured against.
What is not confirmed yet
Everything above rests on the vendor's own post. No independent source, model card, or pricing page appears in the material. The 1M context window, native multimodality, and 552B parameter count are unverified specifications, and the agentic comparison plus the one-third figure are unverified marketing claims. Treat the numbers as a starting point for evaluation rather than settled facts: run the model on your own agentic tasks, check actual per-token pricing, and confirm context limits before migrating any workload.
Source: https://x.com/togethercompute/status/2098940684044562506
Top comments (0)