Terminal-Bench

Summary: DeepSeek V4.1 Flash in 30 Seconds DeepSeek-V4.1-Flash shipped on September 10, 2026. License is MIT, weights are on Hugging Face. 552B backbone parameters, but only 8B active while reading input and 16B while generating output. That asymmetry is the whole point of the model. New architecture: Causal Encoder-Decoder (CED). 40 layers, 20 causal encoder plus 20 decoder. The decoder’s global KV cache is projected from the encoder’s final hidden states instead of from each decoder layer’s own. Result: a global KV cache of 890 bytes per token. One quarter of V4-Flash, and 1/437 of DeepSeek-V1. Natively multimodal: images go through DeepSeek-ViT, a vision encoder trained from scratch, from the very start of language-model pre-training. Pre-training corpus is 45T tokens. Beats Claude Opus 5 on several agentic benchmarks: 90.6 vs 89.1 on Terminal-Bench 2.1, 74.2 vs 74.0 on DeepSWE v1.1, 54.8 vs 50.3 on AutomationBench. Reasoning effort is a continuous 1-100 dial, not an on/off thinking toggle. API pricing is $0.15 in / $0.60 out per million tokens off-peak, double that at peak. Roughly one twentieth of Claude Opus 5’s output price. DeepSeek is retiring V4 Pro: from September 14, 2026, deepseek-v4-pro requests get routed to V4.1 Flash and billed at Flash rates. Everyone in the open-weight race this year is chasing the same two numbers: active parameters and KV cache. The first decides what each token costs, the second decides how much memory a long context eats. GLM-5.3-Flash answered with hybrid attention, Qwen3.8-Flash-Next with 6B active parameters. ...

Meta shipped Muse Spark 1.3 on September 2, 2026: a closed multimodal reasoning model built for long-running agentic workflows, multi-agent setups and coding. Two numbers made the headlines. It scores 75.4 on DeepSWE v1.1, edging past Claude Opus 5, and it hits 98.1% retrieval across a full million tokens of context. Both are real. Both come with a footnote, and the footnotes are what this post is about. What Is Muse Spark 1.3? 🧠 Muse Spark is Meta Superintelligence Labs’ closed flagship series. Version 1.3 succeeds 1.2, and the pitch is not producing one good answer but carrying a long task all the way to the end. ...

Two models, one price tag: $10 per million input tokens, $50 per million output tokens. GPT-6 Astra shipped on September 3, Claude Fable 5.1 on September 1. Identical sticker prices, so the comparison looks simple. It is not. The sticker is the same, the bill is not. Run the same work through both and what you pay can double or halve depending on the shape of your workload. This post starts with where that gap comes from, then walks the benchmark table, and finishes with three concrete billing scenarios. ...

TL;DR: Gemini 3.8 Flash in 30 seconds Gemini 3.8 Flash landed on September 2, 2026, three weeks after 3.7 Flash. That is the third Flash release in three months. Price did not move: $0.75 input, $3.75 output per million tokens. The promo ends December 31, 2026, then it doubles. A second model shipped alongside it: Gemini 3.8 Flash Cyber, tuned for vulnerability discovery and locked behind the new Fairwind Program. The benchmark table is split. It tops the chart on finance, legal, long video and chart reasoning, and it trails Claude Opus 5 badly on long horizon terminal and computer use work. The model “works harder”: more reasoning steps, more iterative tool calls. Same sticker price, potentially a bigger bill. The Flash release cadence has stopped being funny. 3.6 Flash shipped on July 21, 3.7 Flash on August 13, and Gemini 3.8 Flash today, September 2, 2026. Three releases in three months. ...

On September 1, 2026, Anthropic released Claude Fable 5.1 and Claude Mythos 5.1. It has been roughly three months since Claude Fable 5 launched, and this time there is no government directive and no surprise shutdown in the story. The headline is different: the sticker price did not move, but the cost of actually running the model dropped sharply. On top of that, Fable 5.1 beats last month’s Claude Opus 5 on every benchmark Anthropic published. Let’s look at the details. ...