Completed reviews & decisions

Fiction Prose

Assessed Oct 26, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 26, 2026

The model placed 3 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Native route repaired in version172. Saved fiction and screenwriting preliminary screens completed at fourth and third respectively; neither clears the top-two gate. Continue independent writing-overall/business and exact Codex configuration comparisons with original observation dates.

Fiction Prose

Assessed Oct 26, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 28, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Both saved writing screens placed fourth and did not clear the top-two full-test gate. Preserve all tests; continue only independent exact rec4-bit coding comparisons.

Fiction Prose

Assessed Oct 19, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 19, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

October19 Fiction31 and Screenwriting32 completed against the category leaders, both fourth. Tested released October19 checkpoint through native MiniMax :free OpenRouter route at provider-default Max reasoning, unchanged prompts and output budget. Screenwriting required bounded streaming after a nonstreaming timeout. Current writing recommendations retained; no repeated completed screens or full-test approval. Native Claude Code Max coding configuration47873 still requires independent matching evidence; writing outcomes do not transfer.

Fiction Prose

Assessed Oct 21, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 21, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Both saved screens completed October21 after the verified-host repair, placing fourth of four in Fiction Prose and Screenwriting. Retain leaders in those scoped preliminary screens; no full-test approval requested. Continue exact agent/effort/host coding eligibility and independent comparisons separately.

Fiction Prose

Assessed Oct 19, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 19, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

October19 saved screens completed using exact DeepSeek1017 via Morph FP8 with provider-default reasoning, unchanged prompts and preserved leader drafts/ballots. Both placed fourth and did not clear top-two full-test gate. Scoped writing conclusion only; coding Harness Max/OpenCode Max/High still need independent matched evidence. Native DeepSeek host results are not covered by these screens.

Fiction Prose

Assessed Oct 16, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 16, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Writing conclusion applies to meta/muse-spark-open served by Modal in NVFP4 at provider-default reasoning, with unchanged category prompts. Both completed preliminary screens placed fourth; retain the current Fiction Prose and Screenwriting leaders. Do not repeat these completed screens. Coding remains separate: verify exact agent/runtime/effort/quantization and gather matching independent comparisons before any coding conclusion. BF16 or other quantizations are not covered by these writing screens.

Fiction Prose

Assessed Oct 18, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 18, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Both saved writing screens completed after the owner-funded recovery, placing fourth of four in their respective categories without serious judge disagreement. Retain the current writing leaders; do not rerun these completed screens. Continue exact-configuration independent coding comparisons in the separate category follow-up queue; writing outcomes do not establish coding quality.

Fiction Prose

Assessed Oct 12, 2026

The model placed 3 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses. The judges substantially disagreed, so this result deserves particular caution.

Screenwriting

Assessed Oct 12, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Writing screening is complete. Coding remains under review: seek independent comparisons matching Muse Code’s default configuration. Revisit writing if material new evidence or a changed version warrants it; preserve the judges’ disagreement in Fiction Prose.

Fiction Prose

Assessed Oct 9, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 9, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Writing screening is complete. Seek a second qualified coding comparison at matching effort settings and task-specific business-writing evidence. Revisit the writing decision only for material new evidence or a changed version.

Fiction Prose

Assessed Oct 10, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 10, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Writing screening is complete. For coding, verify the usable agent and effort settings and gather independent comparisons of that exact configuration. Revisit writing for material new evidence or a changed release.

Fiction Prose

Assessed Oct 18, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 18, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Both saved writing screens completed after the owner-funded recovery, placing fourth of four in their respective categories without serious judge disagreement. Retain the current writing leaders; do not rerun these completed screens. Continue exact-configuration independent coding comparisons in the separate category follow-up queue; writing outcomes do not establish coding quality.

Image Editing

Assessed Oct 29, 2026

Exact FLUX3Image now has an AA editing entry below the front group. Commercial pixel-preservation tests favor it on local edits, insufficient for broad displacement.

Limits: No BMN media tests. AA snapshot observedOctober29; Arena datedOctober23. Mixed settings, overlapping uncertainty and conflicting second-place direction retained. Commercial Fuser/Morphed/Quantslant/Voyager diagnostics are supplementary, not independent votes. No preview text-to-image eligibility or universal preservation claim.

FLUX 3 Image is a useful precision editor: the original five-edit test found strong preservation of the untouched scene. The accepted Image Editing assessment retained Sunburst, Flare and MAI Image 2.7 Preview. Continue to compare generation, photographs and artistic range separately; this decision does not cover Klein.

Fiction Prose

Assessed Oct 13, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 13, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Writing screening is complete. Coding research continues for Antigravity at Medium and Copilot CLI at default settings. Seek comparisons matching each setup; High-effort results cannot stand in for them. Revisit writing for material new evidence or a changed release.

Capabilities

Assessed Oct 29, 2026

A substantive initial assessment: strong demonstrated reference identity and camera choreography, but not yet our third broad-toolkit recommendation.

Limits: Limited initial editorial assessment from original reports and benchmarks, not BMN-generated tests. Commercial single-run findings supplement two independent publishers; no direct Kling/Wan/Max test. Audio/silent pools, resolutions and modes differ. Source dates retained; missing evidence is not a loss.

Initial limited-confidence assessments are published: Image-to-Video second (proposal113), Text-to-Video second (114), Video Overall third (115); Capabilities assessed outside the podium (116). Existing category queues retain two-day reassessments for independent matched comparisons and repeated success. Remaining unfinished category work: verify actual full Kling 4 footage-edit outputs against existing editing leaders; advertised functions and generation tests do not establish editing quality. Preserve official-source checkedAt October28; this conclusion is not a new release-page check.

Fiction Prose

Assessed Sep 29, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Fiction Prose screening concluded fourth of four; preserve the saved result. Screenwriting full-test work and owner publication gates remain independent and unchanged. Seek a second qualified independent Claude Code High comparison for coding; do not transfer mixed-effort vendor scores.

Image Editing

Assessed Oct 21, 2026

Exact2.1 editing rows reviewed separately from Pro/Turbo; neither supports displacement of selected editors.

Limits: No BMN media tests performed. External pools use different settings; source snapshots retain dates. Overlapping third-place uncertainty, missing surgical/repeated-edit trials and unresolved Krea/nativeKlein mappings remain.

Exact Qwen-Image-2.1 was assessed in the accepted October21 Image Editing comparison and did not displace Sunburst, Flare or Nano Banana 2.1. This is a scoped editorial nonselection, not a verdict on all image tasks. Continue independent generation, photorealism and artistry follow-ups; preserve research-license and local-configuration limitations.

Fiction Prose

Assessed Oct 10, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Screenwriting

Assessed Oct 10, 2026

The model placed 4 of 4 in BMN’s preliminary blind screen against the category leaders at the time. It did not meet the top-two threshold for a full face-off; the published recommendations were retained.

Limits: This is a preliminary AI-panel screen, not a full owner-approved review or a verdict on other uses.

Writing screening is complete. Coding needs a verified agent configuration and inspectable independent comparisons; the existing Rust example lacks enough settings and artifacts. Revisit writing for material new evidence or a changed release.