EsoLang-Bench: Testing Whether LLMs Actually Reason or Just Memorize

Ockhamareto benchmarked against HumanEval / Shared entity: HumanEval / What happened next / Tension

Reinforcement-learning reward shrinks test suites, lifts mutation scores

Linked by a graph relationship (Ockhamareto benchmarked against HumanEval); both cover HumanEval; picks up the HumanEval thread on 2026-08-26.

Shared entities / Earlier coverage

Francois Chollet: ARC-AGI-3 Launches March 25

Both cover Bench, EsoLang, SWE; earlier Bench coverage from 2026-03-11.

Shared entities / Shared topic / What happened next

Vortex makes sparse attention programmable for long-generation agents.

Both cover Directly, LLMs; overlapping topics (directly, llms); picks up the Directly thread on 2026-06-05.

Shared entities / Shared topic / Earlier coverage

DeepSeek V4 Imminent

Both cover HumanEval, SWE; overlapping topics (designed, humaneval); earlier HumanEval coverage from 2026-03-04.

Francois Chollet: ARC-AGI-3 Launching March 25

Both cover Directly, LLMs; overlapping topics (challeng, directly); earlier Directly coverage from 2026-02-12.

Shared entities / What happened next / Tension

Ornith-1.5 ships 397B, 35B and 9B open weights under MIT and lands within a point of Opus 4.8 on Terminal-Bench.

Both cover Bench, SWE; picks up the Bench thread on 2026-08-21; pushes against this story (against).

Argus beats a direct coding agent 78% vs 59% on SWE-Bench Pro at 1.41x tokens, and gets cheaper as it matures.

Both cover Bench, SWE; picks up the Bench thread on 2026-08-06; pushes against this story (vs).

Grok Build Costs $2.49 a Task, Opus 4.8 Costs $11.80

Both cover Bench, SWE; picks up the Bench thread on 2026-07-14; pushes against this story (versus).