Tycho Labs
976 posts
Tycho Labs
@TychoLabsCom
Building trusted AI. Reaching beyond.
Cursor Composer 2.5's is 3–18x cheaper than Opus 4.7 in Claude Code (medium reasoning), and 5–32x cheaper than GPT-5.5 in Codex (medium) based on API pricing This low Cost per Task isn't just driven by relatively low token pricing, it's also driven by low relatively low token usage compared to other leading models. @cursor_ai Composer 2.5 only used 1.6M token to complete our Coding Agent Index benchmarks, while other models used up to 5.7M. This lower token usage also contributes to a low Time per Task. Across the Coding Agent Index configurations shown, average Time per Task was ~12 minutes. Composer 2.5 completed tasks in ~9 minutes on average, making it ~1.3x faster than average, while Composer 2.5 Fast completed tasks in ~7 minutes, making it ~1.8x faster than the average across agents. Link to full benchmark results below
Introducing Agora-1, a multi-agent world model. Multiple participants—human or AI—can now interact inside the same world simulation, all in real-time. Try our playable research preview today, with Agora-1 simulating a multiplayer GoldenEye deathmatch!
update: qwen 3.6 27b dense q4 just one shotted octopus invaders game on a single 3090. hermes agent drove the whole thing, ~41 tok/s gen 21gb vram at full 262k context, thinking mode on. one prompt in and the canonical multi-file space shooter benchmark out, the same exact prompt i ran on qwen 3.5 27b dense back in march on the same card. 3.5 needed one external scope bug fix before the game would even load on first play. 3.6 needed nothing. 11 of 11 files written, 2411 lines of code, zero steering interventions, zero external fixes, playable on first load. 16 minutes 41 seconds wall clock from prompt to playable. consumer tier king on a single 3090 is locked tonight, and the silicon underneath my desk did not change between march and now. the open source ecosystem just moved the floor. watch it ship itself, the full 16 minutes 41 seconds sped to 3 minutes 45, no human touched the keyboard between the first prompt and the final frame.
3 weeks since ml-intern launched and we just hit 1M messages exchanged. that's 3.3 agent-years of ML research in 21 days. 2 months worth of research every day. 17,383 training jobs total. talk about AI acceleration. here's some of what people built: @cmpatino_ replicated the full DeepSeek v4 architecture and pre+post trained a 100M MoE from scratch. → huggingface.co/cmpatino/nanow… it landed a third place submission on @kellerjordan0 optimizer competition. autoresearch on SOTA territory. github.com/KellerJordan/m… @_lewtun Got the intern to convert @AlecRad's cool new talkie-lm 1930 model to work with transformers. tokenizer, chat template, model conversion etc all one-shotted by ml-intern. huggingface.co/lewtun/talkie-… someone created entire PhD dissertation chapter on context-aware agentic cyber defense drafted with 16 research subagents. and someone used it to crack an @Anthropic kernel optimization take-home. (we don't know how to feel about this one 👀 ) just getting started → huggingface.co/spaces/smolage…
Why We Must Build World Models
Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. @subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.
