Hugging Face's logo

Spaces:
taagarwa
/
coding-agent-leaderboard
Running

App Files Community
4
taagarwa commited on
Commit
7bb515c
·
1 Parent(s): d4866e1

🐛 Fix OSS designation for gpt and claude models

Browse files
Files changed (8) hide show
  1. results/swe-bench-pro--ansible-claude-opus-4-8-claude-code.json +1 -1
  2. results/swe-bench-pro--ansible-claude-opus-4-8-opencode.json +1 -1
  3. results/swe-bench-pro--ansible-claude-sonnet-4-6-claude-code.json +1 -1
  4. results/swe-bench-pro--ansible-gpt-5-5-codex.json +1 -1
  5. results/swe-bench-verified-claude-opus-4-8-claude-code.json +1 -1
  6. results/swe-bench-verified-claude-opus-4-8-opencode.json +1 -1
  7. results/swe-bench-verified-claude-sonnet-4-6-claude-code.json +1 -1
  8. results/swe-bench-verified-gpt-5-5-codex.json +1 -1
results/swe-bench-pro--ansible-claude-opus-4-8-claude-code.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
- "is_oss": true,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
 
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
+ "is_oss": false,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
results/swe-bench-pro--ansible-claude-opus-4-8-opencode.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
- "is_oss": true,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
 
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
+ "is_oss": false,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
results/swe-bench-pro--ansible-claude-sonnet-4-6-claude-code.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "Sonnet 4.6",
16
  "repo": "Sonnet 4.6",
17
- "is_oss": true,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-sonnet-4-6"
 
14
  "model": {
15
  "name": "Sonnet 4.6",
16
  "repo": "Sonnet 4.6",
17
+ "is_oss": false,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-sonnet-4-6"
results/swe-bench-pro--ansible-gpt-5-5-codex.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "GPT 5.5 - high",
16
  "repo": "GPT 5.5 - high",
17
- "is_oss": true,
18
  "num_params": 9700,
19
  "precision": "bf16",
20
  "url": "https://openai.com/index/introducing-gpt-5-5"
 
14
  "model": {
15
  "name": "GPT 5.5 - high",
16
  "repo": "GPT 5.5 - high",
17
+ "is_oss": false,
18
  "num_params": 9700,
19
  "precision": "bf16",
20
  "url": "https://openai.com/index/introducing-gpt-5-5"
results/swe-bench-verified-claude-opus-4-8-claude-code.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
- "is_oss": true,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
 
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
+ "is_oss": false,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
results/swe-bench-verified-claude-opus-4-8-opencode.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
- "is_oss": true,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
 
14
  "model": {
15
  "name": "Opus 4.8",
16
  "repo": "Opus 4.8",
17
+ "is_oss": false,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-opus-4-8"
results/swe-bench-verified-claude-sonnet-4-6-claude-code.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "Sonnet 4.6",
16
  "repo": "Sonnet 4.6",
17
- "is_oss": true,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-sonnet-4-6"
 
14
  "model": {
15
  "name": "Sonnet 4.6",
16
  "repo": "Sonnet 4.6",
17
+ "is_oss": false,
18
  "num_params": 1000,
19
  "precision": "bf16",
20
  "url": "https://www.anthropic.com/news/claude-sonnet-4-6"
results/swe-bench-verified-gpt-5-5-codex.json CHANGED
@@ -14,7 +14,7 @@
14
  "model": {
15
  "name": "GPT 5.5 - high",
16
  "repo": "GPT 5.5 - high",
17
- "is_oss": true,
18
  "num_params": 9700,
19
  "precision": "bf16",
20
  "url": "https://openai.com/index/introducing-gpt-5-5"
 
14
  "model": {
15
  "name": "GPT 5.5 - high",
16
  "repo": "GPT 5.5 - high",
17
+ "is_oss": false,
18
  "num_params": 9700,
19
  "precision": "bf16",
20
  "url": "https://openai.com/index/introducing-gpt-5-5"