This page tracks AI model API pricing for coding in 2026: input, cached input and output prices per million tokens for 6 closed frontier models and the leading open-weight models. Every row links to the source we checked. Free to cite with a link back.
AI Model API Prices per Million Tokens
| Model | Input | Cached input | Output | Context | Notes |
|---|---|---|---|---|---|
| GPT-6 Astra OpenAI | $10 | $1 | $50 | 1.05M | >272K-token prompts billed 2x input / 1.5x output · source |
| Claude Opus 5.5 Anthropic | $4 | $0.2 | $20 | — | Fast mode $8 / $40 · source |
| Kimi K3 Moonshot AI (via Together AI) | $2.7 | $0.27 | $13.5 | — | Open weights; Together AI serverless price · source |
| Claude Sonnet 5.5 Anthropic | $2 | $0.2 | $10 | 1M | 128K max output · source |
| GPT-6.1 Sol OpenAI | $2 | $0.1 | $10 | 1.05M | Default model in Codex CLI · source |
| GPT-6 Sol OpenAI | $2 | $0.2 | $10 | — | Superseded by GPT-6.1 Sol · source |
| Gemini 4 Argon Google · limited access | $2 | $0.1 | $10 | 2M | Intro price; rises to $4 / $20. Fairwind Program only · source |
| GLM 5.3 Zhipu AI (via Together AI) | $1.4 | $0.26 | $4.4 | — | Open weights; Together AI serverless price · source |
| Mistral Large 4 Mistral AI · limited access | $1.36 | — | $4.18 | ~520K | 1T total / 49B active MoE; weights due end of Oct 2026 · source |
| DeepSeek V4 Pro DeepSeek (via Together AI) | $1.32 | $0.13 | $3.96 | — | Open weights; Together AI serverless price (V4 Pro 0813) · source |
| DeepSeek V4.1 Flash DeepSeek (via Together AI) | $0.3 | $0.006 | $1.2 | — | Open weights; Together AI serverless price · source |
| GPT-6 Luna OpenAI | $0.1 | — | $0.5 | — | Up to 90% off cached input · source |
| GLM 5.3 Flash Zhipu AI (via Together AI) | $0.15 | $0.03 | $0.5 | — | Open weights; Together AI serverless price · source |
(Prices in USD per 1 million tokens, standard API tier. Open-weight models are priced on Together AI's serverless API. Last verified October 7, 2026.)
Cost of One Agentic Coding Session by Model
Per-token prices hide what you actually pay, because coding agents re-send the same context every turn. This chart prices one illustrative agent session: 10 million input tokens, 90% served from cache, and 300,000 output tokens. It ignores cache-write fees and differences in how many tokens each model uses, so treat it as a comparison, not a quote.
- GPT-6 Astra $34.00
- Claude Opus 5.5 $11.80
- Kimi K3 $9.18
- Claude Sonnet 5.5 $6.80
- GPT-6 Sol $6.80
- GPT-6.1 Sol $5.90
- Gemini 4 Argon (limited) $5.90
- GLM 5.3 $5.06
- Mistral Large 4 (limited) $3.84
- DeepSeek V4 Pro $3.68
- DeepSeek V4.1 Flash $0.71
- GLM 5.3 Flash $0.57
- GPT-6 Luna $0.34
Source: The Vibelog calculation from the prices above
The headline input price is a poor guide on its own. Cache-read pricing decides most of the bill for agent work, which is why GPT-6.1 Sol and Claude Sonnet 5.5 land close together despite different cache prices. For the quality side of the trade-off, see our Claude Opus 5.5 vs GPT-6 coding comparison and the Gemini 4 Argon breakdown.
How We Verify AI Model Prices
Every price comes from the vendor's launch post or pricing page, or from at least two independent outlets reporting the same number. We update this table when a model launches or a price changes, and the date at the top shows the last check. If you spot a stale price, the source link on each row is the fastest way to confirm it.