From 13af2597350e472841e186c05cd4e8296d511426 Mon Sep 17 00:00:00 2001
From: Bakhtiar <bakhtiar27.id@gmail.com>
Date: Tue, 23 Jun 2026 00:20:07 +0700
Subject: [PATCH 1/3] Refresh Neuralwatt GLM catalog and cache pricing

---
 providers/neuralwatt/README.md                |  8 +++---
 .../models/Qwen/Qwen3.5-397B-A17B-FP8.toml    |  3 ++-
 .../models/Qwen/Qwen3.6-35B-A3B.toml          |  3 ++-
 providers/neuralwatt/models/glm-5-fast.toml   | 21 ----------------
 providers/neuralwatt/models/glm-5.1-fast.toml | 21 ----------------
 providers/neuralwatt/models/glm-5.2-fast.toml | 22 ++++++++++++++++
 .../neuralwatt/models/glm-5.2-short-fast.toml | 22 ++++++++++++++++
 .../neuralwatt/models/glm-5.2-short.toml      | 23 +++++++++++++++++
 providers/neuralwatt/models/glm-5.2.toml      |  5 ++--
 .../neuralwatt/models/kimi-k2.5-fast.toml     |  3 ++-
 .../neuralwatt/models/kimi-k2.6-fast.toml     |  5 ++--
 .../models/moonshotai/Kimi-K2.5.toml          |  3 ++-
 .../models/moonshotai/Kimi-K2.6.toml          |  3 ++-
 .../models/moonshotai/Kimi-K2.7-Code.toml     |  3 ++-
 .../neuralwatt/models/qwen3.5-397b-fast.toml  |  3 ++-
 .../neuralwatt/models/qwen3.6-35b-fast.toml   |  3 ++-
 .../models/zai-org/GLM-5.1-FP8.toml           | 25 -------------------
 17 files changed, 93 insertions(+), 83 deletions(-)
 delete mode 100644 providers/neuralwatt/models/glm-5-fast.toml
 delete mode 100644 providers/neuralwatt/models/glm-5.1-fast.toml
 create mode 100644 providers/neuralwatt/models/glm-5.2-fast.toml
 create mode 100644 providers/neuralwatt/models/glm-5.2-short-fast.toml
 create mode 100644 providers/neuralwatt/models/glm-5.2-short.toml
 delete mode 100644 providers/neuralwatt/models/zai-org/GLM-5.1-FP8.toml

diff --git a/providers/neuralwatt/README.md b/providers/neuralwatt/README.md
index f4b00c063..8a2748594 100644
--- a/providers/neuralwatt/README.md
+++ b/providers/neuralwatt/README.md
@@ -12,15 +12,16 @@ Model Categories
 
 Reasoning Models (with interleaved thinking):
 - glm-5.2 — GLM 5.2, reasoning enabled
-- zai-org/GLM-5.1-FP8 — GLM 5.1 FP8, reasoning enabled
+- glm-5.2-short — GLM 5.2 Short, reasoning enabled
 - moonshotai/Kimi-K2.5 — Kimi K2.5, reasoning + image input
 - moonshotai/Kimi-K2.6 — Kimi K2.6, reasoning + image input
+- moonshotai/Kimi-K2.7-Code — Kimi K2.7 Code, reasoning + image input
 - Qwen/Qwen3.5-397B-A17B-FP8 — Qwen3.5 397B, reasoning enabled
 - Qwen/Qwen3.6-35B-A3B — Qwen3.6 35B A3B, reasoning enabled
 
 Fast Variants (optimized for speed, non-reasoning):
-- glm-5-fast — GLM 5 Fast
-- glm-5.1-fast — GLM 5.1 Fast
+- glm-5.2-fast — GLM 5.2 Fast
+- glm-5.2-short-fast — GLM 5.2 Short Fast
 - kimi-k2.5-fast — Kimi K2.5 Fast, image input
 - kimi-k2.6-fast — Kimi K2.6 Fast, image input
 - qwen3.5-397b-fast — Qwen3.5 397B Fast
@@ -28,6 +29,7 @@ Fast Variants (optimized for speed, non-reasoning):
 
 Notes
 - Model IDs, pricing, and limits sourced directly from the Neuralwatt API
+- Cache reads are billed at 25% of the input token price; there is no separate cache-write charge
 - Neuralwatt provides real-time energy consumption data (Joules/kWh) per request
 - "Fast" variants are optimized for lower latency without reasoning
 - Vision models support image input via OpenAI-compatible API
diff --git a/providers/neuralwatt/models/Qwen/Qwen3.5-397B-A17B-FP8.toml b/providers/neuralwatt/models/Qwen/Qwen3.5-397B-A17B-FP8.toml
index bd1e1bcb0..6231221a1 100644
--- a/providers/neuralwatt/models/Qwen/Qwen3.5-397B-A17B-FP8.toml
+++ b/providers/neuralwatt/models/Qwen/Qwen3.5-397B-A17B-FP8.toml
@@ -12,6 +12,7 @@ open_weights = true
 [cost]
 input = 0.69
 output = 4.14
+cache_read = 0.1725
 
 [limit]
 context = 262_128
@@ -22,4 +23,4 @@ input = ["text"]
 output = ["text"]
 
 [interleaved]
-field = "reasoning_content"
+field = "reasoning_content"
\ No newline at end of file
diff --git a/providers/neuralwatt/models/Qwen/Qwen3.6-35B-A3B.toml b/providers/neuralwatt/models/Qwen/Qwen3.6-35B-A3B.toml
index a28e141c1..74ee5e67b 100644
--- a/providers/neuralwatt/models/Qwen/Qwen3.6-35B-A3B.toml
+++ b/providers/neuralwatt/models/Qwen/Qwen3.6-35B-A3B.toml
@@ -12,6 +12,7 @@ open_weights = true
 [cost]
 input = 0.29
 output = 1.15
+cache_read = 0.0725
 
 [limit]
 context = 131_056
@@ -22,4 +23,4 @@ input = ["text", "image"]
 output = ["text"]
 
 [interleaved]
-field = "reasoning_content"
+field = "reasoning_content"
\ No newline at end of file
diff --git a/providers/neuralwatt/models/glm-5-fast.toml b/providers/neuralwatt/models/glm-5-fast.toml
deleted file mode 100644
index 9affa7989..000000000
--- a/providers/neuralwatt/models/glm-5-fast.toml
+++ /dev/null
@@ -1,21 +0,0 @@
-name = "GLM 5 Fast"
-family = "glm"
-release_date = "2026-04-07"
-last_updated = "2026-04-07"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = true
-
-[cost]
-input = 1.1
-output = 3.6
-
-[limit]
-context = 202_736
-output = 202_736
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/neuralwatt/models/glm-5.1-fast.toml b/providers/neuralwatt/models/glm-5.1-fast.toml
deleted file mode 100644
index 797b4d388..000000000
--- a/providers/neuralwatt/models/glm-5.1-fast.toml
+++ /dev/null
@@ -1,21 +0,0 @@
-name = "GLM 5.1 Fast"
-family = "glm"
-release_date = "2026-04-07"
-last_updated = "2026-04-07"
-attachment = false
-reasoning = false
-temperature = true
-tool_call = true
-open_weights = true
-
-[cost]
-input = 1.1
-output = 3.6
-
-[limit]
-context = 202_736
-output = 202_736
-
-[modalities]
-input = ["text"]
-output = ["text"]
diff --git a/providers/neuralwatt/models/glm-5.2-fast.toml b/providers/neuralwatt/models/glm-5.2-fast.toml
new file mode 100644
index 000000000..0df6ffce3
--- /dev/null
+++ b/providers/neuralwatt/models/glm-5.2-fast.toml
@@ -0,0 +1,22 @@
+name = "GLM 5.2 Fast"
+family = "glm"
+release_date = "2026-06-17"
+last_updated = "2026-06-17"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 1.45
+output = 4.5
+cache_read = 0.3625
+
+[limit]
+context = 1_048_560
+output = 1_048_560
+
+[modalities]
+input = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/glm-5.2-short-fast.toml b/providers/neuralwatt/models/glm-5.2-short-fast.toml
new file mode 100644
index 000000000..186d8b9dc
--- /dev/null
+++ b/providers/neuralwatt/models/glm-5.2-short-fast.toml
@@ -0,0 +1,22 @@
+name = "GLM 5.2 Short Fast"
+family = "glm"
+release_date = "2026-06-17"
+last_updated = "2026-06-17"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 1.45
+output = 4.5
+cache_read = 0.3625
+
+[limit]
+context = 199_984
+output = 199_984
+
+[modalities]
+input = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/glm-5.2-short.toml b/providers/neuralwatt/models/glm-5.2-short.toml
new file mode 100644
index 000000000..b9b7461e5
--- /dev/null
+++ b/providers/neuralwatt/models/glm-5.2-short.toml
@@ -0,0 +1,23 @@
+name = "GLM 5.2 Short"
+family = "glm"
+release_date = "2026-06-17"
+last_updated = "2026-06-17"
+attachment = false
+reasoning = true
+reasoning_options = [{ type = "effort", values = ["high", "max"] }]
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 1.45
+output = 4.5
+cache_read = 0.3625
+
+[limit]
+context = 199_984
+output = 199_984
+
+[modalities]
+input = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/glm-5.2.toml b/providers/neuralwatt/models/glm-5.2.toml
index ab16bdf5d..a3a4a7222 100644
--- a/providers/neuralwatt/models/glm-5.2.toml
+++ b/providers/neuralwatt/models/glm-5.2.toml
@@ -4,7 +4,7 @@ release_date = "2026-06-17"
 last_updated = "2026-06-17"
 attachment = false
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["minimal", "low", "medium", "high", "xhigh"] }]
+reasoning_options = [{ type = "effort", values = ["high", "max"] }]
 temperature = true
 tool_call = true
 open_weights = true
@@ -12,6 +12,7 @@ open_weights = true
 [cost]
 input = 1.45
 output = 4.5
+cache_read = 0.3625
 
 [limit]
 context = 1_048_560
@@ -19,4 +20,4 @@ output = 1_048_560
 
 [modalities]
 input = ["text"]
-output = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/kimi-k2.5-fast.toml b/providers/neuralwatt/models/kimi-k2.5-fast.toml
index 4198b1300..6216454a4 100644
--- a/providers/neuralwatt/models/kimi-k2.5-fast.toml
+++ b/providers/neuralwatt/models/kimi-k2.5-fast.toml
@@ -11,6 +11,7 @@ open_weights = true
 [cost]
 input = 0.52
 output = 2.59
+cache_read = 0.13
 
 [limit]
 context = 262_128
@@ -18,4 +19,4 @@ output = 262_128
 
 [modalities]
 input = ["text", "image"]
-output = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/kimi-k2.6-fast.toml b/providers/neuralwatt/models/kimi-k2.6-fast.toml
index 95d675f5b..438589968 100644
--- a/providers/neuralwatt/models/kimi-k2.6-fast.toml
+++ b/providers/neuralwatt/models/kimi-k2.6-fast.toml
@@ -11,6 +11,7 @@ open_weights = true
 [cost]
 input = 0.69
 output = 3.22
+cache_read = 0.1725
 
 [limit]
 context = 262_128
@@ -18,6 +19,4 @@ output = 262_128
 
 [modalities]
 input = ["text", "image"]
-output = ["text"]
-
-
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/moonshotai/Kimi-K2.5.toml b/providers/neuralwatt/models/moonshotai/Kimi-K2.5.toml
index e6e7e08bd..cb937cdf4 100644
--- a/providers/neuralwatt/models/moonshotai/Kimi-K2.5.toml
+++ b/providers/neuralwatt/models/moonshotai/Kimi-K2.5.toml
@@ -12,6 +12,7 @@ open_weights = true
 [cost]
 input = 0.52
 output = 2.59
+cache_read = 0.13
 
 [limit]
 context = 262_128
@@ -22,4 +23,4 @@ input = ["text", "image"]
 output = ["text"]
 
 [interleaved]
-field = "reasoning_content"
+field = "reasoning_content"
\ No newline at end of file
diff --git a/providers/neuralwatt/models/moonshotai/Kimi-K2.6.toml b/providers/neuralwatt/models/moonshotai/Kimi-K2.6.toml
index c6d8e72f0..c3b29dafe 100644
--- a/providers/neuralwatt/models/moonshotai/Kimi-K2.6.toml
+++ b/providers/neuralwatt/models/moonshotai/Kimi-K2.6.toml
@@ -12,6 +12,7 @@ open_weights = true
 [cost]
 input = 0.69
 output = 3.22
+cache_read = 0.1725
 
 [limit]
 context = 262_128
@@ -22,4 +23,4 @@ input = ["text", "image"]
 output = ["text"]
 
 [interleaved]
-field = "reasoning_content"
+field = "reasoning_content"
\ No newline at end of file
diff --git a/providers/neuralwatt/models/moonshotai/Kimi-K2.7-Code.toml b/providers/neuralwatt/models/moonshotai/Kimi-K2.7-Code.toml
index c3e5b56da..76144b1c1 100644
--- a/providers/neuralwatt/models/moonshotai/Kimi-K2.7-Code.toml
+++ b/providers/neuralwatt/models/moonshotai/Kimi-K2.7-Code.toml
@@ -3,7 +3,8 @@ base_model = "moonshotai/kimi-k2.7-code"
 [cost]
 input = 0.95
 output = 4
+cache_read = 0.2375
 
 [modalities]
 input = ["text", "image"]
-output = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/qwen3.5-397b-fast.toml b/providers/neuralwatt/models/qwen3.5-397b-fast.toml
index 14c1f8ea0..022f15639 100644
--- a/providers/neuralwatt/models/qwen3.5-397b-fast.toml
+++ b/providers/neuralwatt/models/qwen3.5-397b-fast.toml
@@ -11,6 +11,7 @@ open_weights = true
 [cost]
 input = 0.69
 output = 4.14
+cache_read = 0.1725
 
 [limit]
 context = 262_128
@@ -18,4 +19,4 @@ output = 262_128
 
 [modalities]
 input = ["text"]
-output = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/qwen3.6-35b-fast.toml b/providers/neuralwatt/models/qwen3.6-35b-fast.toml
index 53ec205f4..e6d0dcc0b 100644
--- a/providers/neuralwatt/models/qwen3.6-35b-fast.toml
+++ b/providers/neuralwatt/models/qwen3.6-35b-fast.toml
@@ -11,6 +11,7 @@ open_weights = true
 [cost]
 input = 0.29
 output = 1.15
+cache_read = 0.0725
 
 [limit]
 context = 131_056
@@ -18,4 +19,4 @@ output = 131_056
 
 [modalities]
 input = ["text", "image"]
-output = ["text"]
+output = ["text"]
\ No newline at end of file
diff --git a/providers/neuralwatt/models/zai-org/GLM-5.1-FP8.toml b/providers/neuralwatt/models/zai-org/GLM-5.1-FP8.toml
deleted file mode 100644
index 1d35519c0..000000000
--- a/providers/neuralwatt/models/zai-org/GLM-5.1-FP8.toml
+++ /dev/null
@@ -1,25 +0,0 @@
-name = "GLM 5.1 FP8"
-family = "glm"
-release_date = "2026-04-07"
-last_updated = "2026-04-07"
-attachment = false
-reasoning = true
-reasoning_options = [{ type = "toggle" }]
-temperature = true
-tool_call = true
-open_weights = true
-
-[cost]
-input = 1.1
-output = 3.6
-
-[limit]
-context = 202_736
-output = 202_736
-
-[modalities]
-input = ["text"]
-output = ["text"]
-
-[interleaved]
-field = "reasoning_content"

From 3d0f1dd861edb9e54afc0f95f82806c44d2f1fbf Mon Sep 17 00:00:00 2001
From: Bakhtiar <bakhtiar27.id@gmail.com>
Date: Sat, 27 Jun 2026 00:06:40 +0700
Subject: [PATCH 2/3] Add documented GLM-5.2 reasoning controls

Restore the documented reasoning surface for Neuralwatt GLM-5.2 and
GLM-5.2 Short by adding toggle, budget_tokens, and the full supported
effort set.

This also resolves the regression from dev where the models were reduced
to high/max-only effort options.
---
 providers/neuralwatt/models/glm-5.2-short.toml | 6 +++++-
 providers/neuralwatt/models/glm-5.2.toml       | 8 ++++++--
 2 files changed, 11 insertions(+), 3 deletions(-)

diff --git a/providers/neuralwatt/models/glm-5.2-short.toml b/providers/neuralwatt/models/glm-5.2-short.toml
index d75504835..a1aa3c1a2 100644
--- a/providers/neuralwatt/models/glm-5.2-short.toml
+++ b/providers/neuralwatt/models/glm-5.2-short.toml
@@ -4,7 +4,11 @@ release_date = "2026-06-17"
 last_updated = "2026-06-17"
 attachment = false
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["high", "max"] }]
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] },
+  { type = "budget_tokens" },
+]
 temperature = true
 tool_call = true
 open_weights = true
diff --git a/providers/neuralwatt/models/glm-5.2.toml b/providers/neuralwatt/models/glm-5.2.toml
index a3a4a7222..db8a85871 100644
--- a/providers/neuralwatt/models/glm-5.2.toml
+++ b/providers/neuralwatt/models/glm-5.2.toml
@@ -4,7 +4,11 @@ release_date = "2026-06-17"
 last_updated = "2026-06-17"
 attachment = false
 reasoning = true
-reasoning_options = [{ type = "effort", values = ["high", "max"] }]
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] },
+  { type = "budget_tokens" },
+]
 temperature = true
 tool_call = true
 open_weights = true
@@ -20,4 +24,4 @@ output = 1_048_560
 
 [modalities]
 input = ["text"]
-output = ["text"]
\ No newline at end of file
+output = ["text"]

From 286d0328c358a392377bd871d4477c4f8188f541 Mon Sep 17 00:00:00 2001
From: Bakhtiar <bakhtiar27.id@gmail.com>
Date: Sun, 28 Jun 2026 13:51:12 +0700
Subject: [PATCH 3/3] Add 5 flex tier model variants

---
 providers/neuralwatt/README.md                | 11 +++++++-
 providers/neuralwatt/models/glm-5.2-flex.toml | 27 +++++++++++++++++++
 .../models/glm-5.2-short-fast-flex.toml       | 22 +++++++++++++++
 .../neuralwatt/models/glm-5.2-short-flex.toml | 27 +++++++++++++++++++
 .../neuralwatt/models/kimi-k2.6-flex.toml     | 26 ++++++++++++++++++
 .../models/kimi-k2.7-code-flex.toml           | 11 ++++++++
 providers/neuralwatt/provider.toml            |  6 +++++
 7 files changed, 129 insertions(+), 1 deletion(-)
 create mode 100644 providers/neuralwatt/models/glm-5.2-flex.toml
 create mode 100644 providers/neuralwatt/models/glm-5.2-short-fast-flex.toml
 create mode 100644 providers/neuralwatt/models/glm-5.2-short-flex.toml
 create mode 100644 providers/neuralwatt/models/kimi-k2.6-flex.toml
 create mode 100644 providers/neuralwatt/models/kimi-k2.7-code-flex.toml

diff --git a/providers/neuralwatt/README.md b/providers/neuralwatt/README.md
index 8a2748594..e3e2397c5 100644
--- a/providers/neuralwatt/README.md
+++ b/providers/neuralwatt/README.md
@@ -27,9 +27,18 @@ Fast Variants (optimized for speed, non-reasoning):
 - qwen3.5-397b-fast — Qwen3.5 397B Fast
 - qwen3.6-35b-fast — Qwen3.6 35B Fast
 
+Flex Variants (streaming required, discounted):
+- glm-5.2-flex — GLM 5.2 Flex, reasoning enabled
+- glm-5.2-short-flex — GLM 5.2 Short Flex, reasoning enabled
+- glm-5.2-short-fast-flex — GLM 5.2 Short Fast Flex
+- kimi-k2.6-flex — Kimi K2.6 Flex, reasoning + image input
+- kimi-k2.7-code-flex — Kimi K2.7 Code Flex, reasoning + image input
+
 Notes
-- Model IDs, pricing, and limits sourced directly from the Neuralwatt API
+- Standard model IDs, pricing, and limits are sourced directly from the Neuralwatt API; flex pricing applies the official 0.5 docs multiplier to the corresponding standard rates
 - Cache reads are billed at 25% of the input token price; there is no separate cache-write charge
 - Neuralwatt provides real-time energy consumption data (Joules/kWh) per request
 - "Fast" variants are optimized for lower latency without reasoning
+- Flex requires streaming; non-streaming requests fall back to the standard tier
+- Official Neuralwatt docs currently describe Flex as a 50% token-pricing discount, including cached-input billing
 - Vision models support image input via OpenAI-compatible API
diff --git a/providers/neuralwatt/models/glm-5.2-flex.toml b/providers/neuralwatt/models/glm-5.2-flex.toml
new file mode 100644
index 000000000..585304de9
--- /dev/null
+++ b/providers/neuralwatt/models/glm-5.2-flex.toml
@@ -0,0 +1,27 @@
+name = "GLM 5.2 Flex"
+family = "glm"
+release_date = "2026-06-17"
+last_updated = "2026-06-17"
+attachment = false
+reasoning = true
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] },
+  { type = "budget_tokens" },
+]
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 0.725
+output = 2.25
+cache_read = 0.18125
+
+[limit]
+context = 1_048_560
+output = 1_048_560
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/neuralwatt/models/glm-5.2-short-fast-flex.toml b/providers/neuralwatt/models/glm-5.2-short-fast-flex.toml
new file mode 100644
index 000000000..1af1a6507
--- /dev/null
+++ b/providers/neuralwatt/models/glm-5.2-short-fast-flex.toml
@@ -0,0 +1,22 @@
+name = "GLM 5.2 Short Fast Flex"
+family = "glm"
+release_date = "2026-06-17"
+last_updated = "2026-06-17"
+attachment = false
+reasoning = false
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 0.725
+output = 2.25
+cache_read = 0.18125
+
+[limit]
+context = 199_984
+output = 199_984
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/neuralwatt/models/glm-5.2-short-flex.toml b/providers/neuralwatt/models/glm-5.2-short-flex.toml
new file mode 100644
index 000000000..ff1086ad7
--- /dev/null
+++ b/providers/neuralwatt/models/glm-5.2-short-flex.toml
@@ -0,0 +1,27 @@
+name = "GLM 5.2 Short Flex"
+family = "glm"
+release_date = "2026-06-17"
+last_updated = "2026-06-17"
+attachment = false
+reasoning = true
+reasoning_options = [
+  { type = "toggle" },
+  { type = "effort", values = ["none", "minimal", "low", "medium", "high", "xhigh", "max"] },
+  { type = "budget_tokens" },
+]
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 0.725
+output = 2.25
+cache_read = 0.18125
+
+[limit]
+context = 199_984
+output = 199_984
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/neuralwatt/models/kimi-k2.6-flex.toml b/providers/neuralwatt/models/kimi-k2.6-flex.toml
new file mode 100644
index 000000000..ff9963779
--- /dev/null
+++ b/providers/neuralwatt/models/kimi-k2.6-flex.toml
@@ -0,0 +1,26 @@
+name = "Kimi K2.6 Flex"
+family = "kimi-k2"
+release_date = "2026-04-21"
+last_updated = "2026-04-21"
+attachment = true
+reasoning = true
+reasoning_options = [{ type = "toggle" }]
+temperature = true
+tool_call = true
+open_weights = true
+
+[cost]
+input = 0.345
+output = 1.61
+cache_read = 0.08625
+
+[limit]
+context = 262_128
+output = 262_128
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
+
+[interleaved]
+field = "reasoning_content"
diff --git a/providers/neuralwatt/models/kimi-k2.7-code-flex.toml b/providers/neuralwatt/models/kimi-k2.7-code-flex.toml
new file mode 100644
index 000000000..10945afba
--- /dev/null
+++ b/providers/neuralwatt/models/kimi-k2.7-code-flex.toml
@@ -0,0 +1,11 @@
+base_model = "moonshotai/kimi-k2.7-code"
+name = "Kimi K2.7 Code Flex"
+
+[cost]
+input = 0.475
+output = 2
+cache_read = 0.11875
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
diff --git a/providers/neuralwatt/provider.toml b/providers/neuralwatt/provider.toml
index 6e207a1b2..c7784906c 100644
--- a/providers/neuralwatt/provider.toml
+++ b/providers/neuralwatt/provider.toml
@@ -5,6 +5,12 @@ npm = "@ai-sdk/openai-compatible"
 # `thinking_token_budget` (no bounds or disable sentinel documented). Native
 # toggles use `chat_template_kwargs.enable_thinking = true|false`; GLM-5.2 also
 # accepts `reasoning_effort`, with the model-specific normalization documented.
+# Flex tier uses either a `-flex` model ID or `service_tier = "flex"` and
+# requires `stream = true`; non-streaming requests fall through to the standard
+# tier. The official Energy Methodology docs currently state a 0.5 flex pricing
+# multiplier.
 # https://portal.neuralwatt.com/docs/api/chat-completions (accessed 2026-06-25)
+# https://portal.neuralwatt.com/docs/guides/flex-tier
+# https://portal.neuralwatt.com/docs/energy-methodology (accessed 2026-06-28)
 api = "https://api.neuralwatt.com/v1"
 doc = "https://portal.neuralwatt.com/docs"