chore(pricing): Update fireworks-ai pricing

[siddharthsambharia-portkey]

🔄 Pricing Update: fireworks-ai

📊 Summary (complete_diff mode)

Change Type	Count
➕ Models added	17
🔄 Models updated (merged)	4

➕ New Models

• deepseek-v3p1
• deepseek-v3p2
• glm-4p7
• glm-5
• qwen3-vl-30b-a3b-instruct
• qwen3-vl-30b-a3b-thinking
• minimax-m2p1
• minimax-m2p5
• gpt-oss-120b
• gpt-oss-20b
• llama-v3p3-70b-instruct
• qwen3-8b
• flux-1-dev-fp8
• flux-1-schnell-fp8
• flux-kontext-pro
• flux-kontext-max
• qwen3-embedding-8b

🔄 Updated Models

• kimi-k2-instruct-0905
• kimi-k2-thinking
• kimi-k2p5
• mixtral-8x22b-instruct

Pricing Sources

Tier Table (from fireworks.ai/pricing)

Range	Input $/1M	Output $/1M
<4B	$0.10	$0.10
4B–16B	$0.20	$0.20
>16B	$0.90	$0.90
MoE 0–56B	$0.50	$0.50
MoE 56.1–176B	$1.20	$1.20

Named Families

Model	Input	Cached	Output
DeepSeek V3.1 / V3.2	$0.56	$0.28	$1.68
GLM-4.7	$0.60	$0.30	$2.20
GLM-5	$1.00	$0.20	$3.20
Qwen3 VL 30B A3B	$0.15	$0.075	$0.60
Kimi K2 / K2 Thinking	$0.60	$0.30	$2.50
Kimi K2.5	$0.60	$0.10	$3.00
MiniMax M2.1 / M2.5	$0.30	$0.03	$1.20
gpt-oss-120b	$0.15	$0.075	$0.60
gpt-oss-20b	$0.07	$0.035	$0.30

Image Generation

Model	Price
FLUX.1 [dev] FP8	$0.00025/step
FLUX.1 [schnell] FP8	$0.00035/step
FLUX Kontext Pro	$0.04/image
FLUX Kontext Max	$0.08/image

Embeddings

Model	Input $/1M
Qwen3 Embedding 8B	$0.008

Rules Applied

• Cache read = 50% of input (unless named-family specifies otherwise)
• Batch = 50% of serverless input + 50% of serverless output
• No cache_write_price set (Fireworks charges cache read only)
• qwen3-reranker-8b skipped (reranker, not LLM/embed)

---

Generated by Pricing Agent on 2026-03-31