Quantized Qwen3.8 27B for agentic coding: Same Score, 42 Different Outcomes.  I’ve been evaluating quantized Qwen3.8 27B models on DeepSWE 1.1, but I’m giving up. 4-bit AWQ run #1#: 38.94% Reward 4-bit AWQ run #2#: 31.86% Reward 32 of the 113 task outcomes flipped: 20 previously solved tasks failed, while 12 previously failed tasks passed. On non-agentic tasks, I’ve found NVIDIA NVFP4 performs very closely to AWQ. On DeepSWE 1.1? NVIDIA NVFP4: 31.86% Reward Exactly the same score as AWQ run #2#, but 42 task outcomes flipped (nearly one-third of the benchmark!). Each solved 36 tasks, with only 15 successes in common. For reference, BF16 with the same Pi harness and thinking mode reached 43.36%. Quantization certainly contributes to trajectory instability: small changes in token probabilities can alter an edit or tool call. That changes what the agent sees next, potentially sending subsequent decisions down a different path. Errors can compound across turns. Separating that from ordinary sampling variability requires repeated experiments. I’d want at least three independent full runs per model with fixed settings before drawing conclusions, and more for a reliable estimate of variability. Too long, too expensive. Don’t trust anyone reporting a quantized model’s accuracy on agentic tasks without also reporting the run-to-run standard deviation.

Quantifying the qual >> @qualitateio turns expert & tech buyer discussions into quantified signals across markets, companies, and time periods Every number on the platform links back to the underlying conversations Now, you can click into any metric and summarize those conversations in one click Quick look below with @databricks:

QUANTINUUM SIGNS MOU WITH ARAMCO Quantinuum $QNT and Aramco signed a non-binding agreement to explore quantum computing applications across the energy sector. The companies will identify industrial problems that could benefit from quantum computing, benchmark different quantum approaches, and prepare for a potential research collaboration focused on fault-tolerant systems. The work will also include technical onboarding and knowledge sharing around future energy and digital-transformation use cases.

Quantitative Easing swaps a long bond for a central bank reserve that gets stuck on a commercial bank balance sheet earning IORB treasury buybacks swap a long bond for a money like Tbill that is the most liquid and modular collateral that exists Size of operations aside, you can make an argument the current program being at least as equally potent. Especially if Fed were to begin buying those bills, then thats outright debt monetization indirectly. Just some thoughts

Quantic Dream is shutting down the servers for its game Spellcasters Chronicles, which lasted only 83 days. It is shutting down because nobody is playing it. Steam charts showed it peaked at about 888 concurrent players soon after launch, but the numbers dropped fast, with as few as 54 players in recent weeks. The studio will give full refunds to everyone who bought Early Access and any in-game items. The company will also have layoffs while work on their other game Star Wars Eclipse is unaffected and continues as planned.