chore: update berget.ai models - prices and Gemma 4

- Add Google Gemma 4 31B Instruct model
- Update prices for existing models (EUR to USD conversion)
- Remove non-coding models (bge-reranker, multilingual-e5 embeddings, kb-whisper)
- Remove deprecated Llama-3.1-8B-Instruct

Updated models:
- GLM-4.7: 0.77/2.75 USD/M (was 0.7/2.3)
- Llama-3.3-70B: 0.99/0.99 USD/M (was 0.9/0.9)
- Mistral-Small-3.2: 0.33/0.33 USD/M (was 0.3/0.3)
- GPT-OSS-120B: 0.44/0.99 USD/M (was 0.3/0.9)

New models:
- Gemma-4-31B-it: 0.275/0.55 USD/M

Removed models (not relevant for coding):
- BAAI/bge-reranker-v2-m3 (reranker)
- intfloat/multilingual-e5-large/* (embeddings)
- KBLab/kb-whisper-large (speech-to-text)
- meta-llama/Llama-3.1-8B-Instruct (deprecated)

Author: irony

providers/berget/models/BAAI/bge-reranker-v2-m3.toml

@@ -0,0 +1,23 @@
+name = "Gemma 4 31B Instruct"
+family = "gemma"
+release_date = "2026-04-02"
+last_updated = "2026-04-02"
+attachment = false
+reasoning = true
+temperature = true
+knowledge = "2025-12"
+tool_call = true
+structured_output = true
+open_weights = true
+
+[cost]
+input = 0.275
+output = 0.55
+
+[limit]
+context = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]

providers/berget/models/KBLab/kb-whisper-large.toml

@@ -11,8 +11,8 @@ structured_output = true
 open_weights = true
 
 [cost]
-input = 0.9
-output = 0.9
+input = 0.99
+output = 0.99
 
 [limit]
 context = 128_000

providers/berget/models/google/gemma-4-31B-it.toml

@@ -11,8 +11,8 @@ structured_output = true
 open_weights = true
 
 [cost]
-input = 0.3
-output = 0.3
+input = 0.33
+output = 0.33
 
 [limit]
 context = 32_000

providers/berget/models/intfloat/multilingual-e5-large-instruct.toml

@@ -11,8 +11,8 @@ structured_output = true
 open_weights = true
 
 [cost]
-input = 0.3
-output = 0.9
+input = 0.44
+output = 0.99
 
 [limit]
 context = 128_000

providers/berget/models/intfloat/multilingual-e5-large.toml

@@ -11,8 +11,8 @@ structured_output = true
 open_weights = true
 
 [cost]
-input = 0.7
-output = 2.3
+input = 0.77
+output = 2.75
 
 [limit]
 context = 128_000