50%/80% horizon ratio

consistent with normal70/95leading

1What this measures

METR 50% horizon divided by 80% horizon for the same model (derived metric horizon_ratio_80_50). ~7-10x recently, up from ~5x.

Why it matters. The ratio is the reliability gap in one number. If it collapses toward 2x while the 80% horizon exceeds eight hours, the normal-technology thesis loses its productisation argument.

Proxy types: benchmark
Unit: ratio
Cadence: per release
Valve: invention to product

2How we track this

derived metric horizon_ratio_80_50 (formula in the semantic layer)

Normal band: ≥ 5.00×
Fast band: ≤ 2.00×
Falsifying: —

Normal = ratio ≥ 5 (gap holding or widening); fast = ratio ≤ 2. The falsifying condition (≤ 2 AND 80% horizon > 8 h) is compound and lives in the thesis monitor, not the band.

Applied to metric:horizon_ratio_80_50.

3Tracker interpretation

The gap is widening, which is pro-normal. It is the paper's reliability argument made measurable.

4Evidence

Latest point

6.34×as of 2026-03-05(2 obs)

Value the bands apply to

6.34×as of 2026-03-05(2 obs)

50 observations. Hollow points are disputed (see counterevidence). Every point links to its observation.

Derived rows (25)

as of	dims	value	inputs
2026-03-05	gpt_5_4	6.34×	obs:11a8bbd2 obs:858c619b
2026-02-19	gemini_3_1_pro	4.28×	obs:2eb7c897 obs:5eb08488
2026-02-05	gpt_5_3_codex	6.39×	obs:959db6dd obs:e466aae4
2026-02-05	claude_opus_4_6_inspect	10.3×	obs:39c8213a obs:8f73b56b
2025-12-11	gpt_5_2	5.34×	obs:09061051 obs:8a2d6997
2025-11-24	claude_opus_4_5_inspect	5.93×	obs:72bcb9d8 obs:75c93027
2025-11-19	gpt_5_1_codex_max_inspect	4.42×	obs:8de1effa obs:a2aa2e4d
2025-11-18	gemini_3_pro	4.14×	obs:3466aea4 obs:833be5d2
2025-08-07	gpt_5_2025_08_07_inspect	5.30×	obs:261c1f51 obs:47db79bd
2025-08-05	claude_4_1_opus_inspect	4.28×	obs:ae527806 obs:fecd07e2
2025-05-22	claude_4_opus_inspect	4.91×	obs:1f69c372 obs:6af93ac0
2025-04-16	o3_inspect	3.99×	obs:6b19d566 obs:d1bb5ffe
2025-02-24	claude_3_7_sonnet_inspect	4.99×	obs:095bb211 obs:68864ed5
2024-12-05	o1_inspect	5.48×	obs:37e97a6f obs:9450442a
2024-10-22	claude_3_5_sonnet_20241022_inspect	7.91×	obs:0fb6ae10 obs:19631991
2024-09-12	o1_preview	4.60×	obs:734779b5 obs:b5c1303b
2024-06-20	claude_3_5_sonnet_20240620_inspect	6.82×	obs:145115d4 obs:efa58e1c
2024-05-13	gpt_4o_inspect	5.52×	obs:2dc1a032 obs:7c977050
2024-04-09	gpt_4_turbo_inspect	4.02×	obs:0d21fe65 obs:deb6a1ab
2024-03-04	claude_3_opus_inspect	6.19×	obs:f1409c69 obs:fb7435cd
2023-11-06	gpt_4_1106_inspect	5.17×	obs:29446d6c obs:cd50d564
2023-03-14	gpt_4	4.48×	obs:96d016fa obs:a53da98c
2022-03-15	gpt_3_5_turbo_instruct	2.35×	obs:4a682b8c obs:5e6df781
2020-05-28	davinci_002	2.56×	obs:2325ad87 obs:58c5e677
2019-02-14	gpt2	4.20×	obs:1c22fa34 obs:64f0dec4

5Status and reasoning

consistent with normalsince 2026-09-10 · claude-initial-seed

Latest non-disputed model ratio is 6.3x (>= 5 normal band); the Mythos point is excluded as beyond the 16h suite ceiling. Matches the brief's expectation. Initial seed; pending Alex's review.

6Timeline notes

2026-03-05 gpt_5_4 · 6.34×as of 2026-03-05(2 obs)
2026-02-19 gemini_3_1_pro · 4.28×as of 2026-02-19(2 obs)
2026-02-05 gpt_5_3_codex · 6.39×as of 2026-02-05(2 obs)
2026-02-05 claude_opus_4_6_inspect · 10.3×as of 2026-02-05(2 obs)
2025-12-11 gpt_5_2 · 5.34×as of 2025-12-11(2 obs)
2025-11-24 claude_opus_4_5_inspect · 5.93×as of 2025-11-24(2 obs)

7Counterevidence

What cuts against this reading

Ratio is sensitive to the task mix and to the exclusion of models beyond the suite ceiling; a single new model can move it.

8Update history

2026-09-10unmeasured to consistent with normalconf — → 70 · claude-initial-seed
Latest non-disputed model ratio is 6.3x (>= 5 normal band); the Mythos point is excluded as beyond the 16h suite ceiling. Matches the brief's expectation. Initial seed; pending Alex's review.
obs:11a8bbd2 obs:858c619b

9Confidence

70 / 95 — good evidence, some ambiguity

Confidence is independent of status: 90–95 multiple strong independent sources; 70–89 good evidence, some ambiguity; 50–69 mixed or hard to operationalise; below 50 limited or vague.

10Related

METR 50% time horizon faster than normal
METR 80% time horizon faster than normal
Bottlenecks #3, #4, #16, #65 (Narayanan & Kapoor's list)
Crosswalk: Methods ⇄ Model (same valve)