Skip to content
Fifth Order
← Benchmarks

Investment Research · Fifth Order

Thesis Revision

Models revise investment views as new operating evidence arrives across a collection of company cases. Evaluation checks whether assumptions change with the evidence, valuation implications are explicit and counter-evidence receives fair weight. Each revision must name conditions that would invalidate the view. Scores combine these rubric criteria across 40 cases, rewarding analytical quality rather than polished prose alone.

Model results

28 configurations
RankModelScore
  1. 1
    AnthropicClaude Opus 5.5 Adaptive/Max
    88.3%
  2. 2
    AnthropicClaude Fable 5.1 Adaptive/Max
    86.7%
  3. 3
    AnthropicClaude Sonnet 5.5 Adaptive/Max
    85.6%
  4. 4
    OpenAIGPT 6 Astra Max reasoning
    85.1%
  5. 5
    AnthropicClaude Opus 5 Adaptive/Max
    84.5%
  6. 6
    MetaMuse Spark 1.3 Max reasoning
    83.6%
  7. 7
    MetaMuse Wonder 1.0 Max reasoning
    82.2%
  8. 8
    xAIGrok 4.6 xHigh reasoning
    82.1%
  9. 9
    OpenAIGPT 6.1 Sol Max reasoning
    81.9%
  10. 10
    GoogleGemini 4 Argon High reasoning
    80.7%
  11. 11
    TencentHy4 Preview High reasoning
    80.1%
  12. 12
    AnthropicClaude Fable 5 Adaptive/Max
    79.7%
  13. 13
    OpenAIGPT 6 Luna Max reasoning
    79.4%
  14. 14
    xAIGrok 4.7 xHigh reasoning
    79.1%
  15. 15
    AlibabaQwen 3.8 Max xHigh reasoning
    78.9%
  16. 16
    MistralMistral Large 4 High reasoning
    77.6%
  17. 17
    OpenAIGPT 6 Sol Max reasoning
    77.4%
  18. 18
    DeepSeekDeepSeek V4.1 Flash 1017 Max reasoning
    77.3%
  19. 19
    ZhipuGLM 5.3 Flash Max reasoning
    76.8%
  20. 20
    ZhipuGLM 5.3 Max reasoning
    76.2%
  21. 21
    OpenAIGPT 5.6 Sol Max reasoning
    74.6%
  22. 22
    MoonshotKimi K3 Max reasoning
    74.3%
  23. 23
    GoogleGemini 3.8 Flash High reasoning
    74.1%
  24. 24
    AnthropicClaude Sonnet 5 Adaptive/Max
    72.9%
  25. 25
    OpenAIGPT 5.6 Luna Max reasoning
    72.6%
  26. 26
    MetaMuse Spark 1.2 xHigh reasoning
    72.5%
  27. 27
    GoogleGemini 3.7 Flash High reasoning
    71.8%
  28. 28
    DeepSeekDeepSeek V4 Pro Max reasoning
    71.2%
Numeric table · 28 configurations
Rubric score · highest first
RankModel / settingScore
1
AnthropicClaude Opus 5.5 Adaptive/Max
88.3%
2
AnthropicClaude Fable 5.1 Adaptive/Max
86.7%
3
AnthropicClaude Sonnet 5.5 Adaptive/Max
85.6%
4
OpenAIGPT 6 Astra Max reasoning
85.1%
5
AnthropicClaude Opus 5 Adaptive/Max
84.5%
6
MetaMuse Spark 1.3 Max reasoning
83.6%
7
MetaMuse Wonder 1.0 Max reasoning
82.2%
8
xAIGrok 4.6 xHigh reasoning
82.1%
9
OpenAIGPT 6.1 Sol Max reasoning
81.9%
10
GoogleGemini 4 Argon High reasoning
80.7%
11
TencentHy4 Preview High reasoning
80.1%
12
AnthropicClaude Fable 5 Adaptive/Max
79.7%
13
OpenAIGPT 6 Luna Max reasoning
79.4%
14
xAIGrok 4.7 xHigh reasoning
79.1%
15
AlibabaQwen 3.8 Max xHigh reasoning
78.9%
16
MistralMistral Large 4 High reasoning
77.6%
17
OpenAIGPT 6 Sol Max reasoning
77.4%
18
DeepSeekDeepSeek V4.1 Flash 1017 Max reasoning
77.3%
19
ZhipuGLM 5.3 Flash Max reasoning
76.8%
20
ZhipuGLM 5.3 Max reasoning
76.2%
21
OpenAIGPT 5.6 Sol Max reasoning
74.6%
22
MoonshotKimi K3 Max reasoning
74.3%
23
GoogleGemini 3.8 Flash High reasoning
74.1%
24
AnthropicClaude Sonnet 5 Adaptive/Max
72.9%
25
OpenAIGPT 5.6 Luna Max reasoning
72.6%
26
MetaMuse Spark 1.2 xHigh reasoning
72.5%
27
GoogleGemini 3.7 Flash High reasoning
71.8%
28
DeepSeekDeepSeek V4 Pro Max reasoning
71.2%
Methodology & reference example

Version finance-index-v1 · October 1–28, 2026 · 40 cases per component. Four equally weighted rubric criteria produce a 0–100 score: evidence consistency; assumption and valuation updates; counter-evidence; invalidation conditions. The Credo packet below is an authored reference example, not a model run or the source of these cohort scores.

Task instructions

Begin with the evidence in the packet: reported execution and management's outlook. Update the thesis, assumptions, countercase and invalidation conditions, separating evidence from judgment. Keep durability, mix, margin and valuation questions open where the sources do not resolve them.

Credo Technology Group packet

Public results release · June 1, 2026 · historical cutoff June 11, 2026 · USD millions unless stated

Credo FY2026 results · June 1, 2026

Company release · June 1, 2026. Q4 FY2026 (ended May 2, 2026) revenue of $437.0 million, up 7.4% sequentially and up 157.0% year over year. Q4 non-GAAP gross margin 68.3% (GAAP 68.2%). Q4 non-GAAP diluted EPS $1.16 (GAAP $0.88). FY2026 revenue $1,335.116 million versus $436.775 million in FY2025. Q1 FY2027 outlook: revenue $465–475 million; non-GAAP gross margin 67–69%; GAAP gross margin 66.9–68.9%; GAAP and non-GAAP operating expenses.

SEC exhibit mirror · June 1, 2026

SEC-filed exhibit mirroring the June 1, 2026 results release. The same reported figures apply: Q4 FY2026 revenue $437.0 million, non-GAAP gross margin 68.3%, non-GAAP diluted EPS $1.16, and Q1 FY2027 guidance covering revenue of $465–475 million, non-GAAP gross margin of 67–69%, GAAP gross margin of 66.9–68.9%, and GAAP and non-GAAP operating expenses.

Authored reference output

Thesis review — execution strong, valuation unanswered

Observed execution supports the growth thesis; the questions that remain are about durability, mix, margins and expectations — and none of them is answered by one more strong quarter alone.

What the evidence supports

Revenue grew 157.0% year over year in Q4 FY2026 and FY2026 revenue roughly tripled to $1,335.116m; Q4 non-GAAP gross margin was 68.3%. Management then guided to a further ~7.55% sequential increase. Execution against the AI-infrastructure demand thesis is strong on the reported record.

Credo FY2026 results · June 1, 2026 · SEC exhibit mirror · June 1, 2026

Remaining questions

Durability: how much of the growth is a concentrated buildout cycle versus a recurring base? Mix: does the copper-to-optical transition hold margins? Margins: the guided 67–69% range brackets Q4's 68.3%; its 68% midpoint is 0.3 percentage points lower — is that conservatism or early pressure? Expectations: the analysis contains no price data, so none of this addresses what the market already pays for the growth.

Credo FY2026 results · June 1, 2026

Evidence that would change the view

Strengthening: sustained sequential growth with margin inside the guided range and broadening customer commentary. Challenging: a sequential miss against $465m, margin below 67%, or disclosure that growth depends on a smaller set of buyers than assumed. Valuation discipline stands separately: even a strengthening business can be a poor entry at the wrong price.

Credo FY2026 results · June 1, 2026 · SEC exhibit mirror · June 1, 2026

Criterion-level review approach

Evidence consistency
The revision must reflect the reported results and guidance without inventing durability, mix detail or market pricing.
Calibrated expert judgment
Distinguish sourced facts from assumptions and conditional judgments; strong reported growth alone does not settle durability or valuation.
Countercase & invalidation
Name the evidence that would strengthen or challenge the view—sequential performance against the guide, margin within the range, customer breadth—without invented probabilities.
Valuation discipline
Keep the valuation question open: this packet contains no price data, and business quality is not an entry-price conclusion.